Disclosure of Invention
Accordingly, the main objective of the present invention is to provide an intelligent report system and method for whole-body tumor MRI evaluation, which can solve the problems of increased examination cost of patients and reduced diagnosis efficiency of clinicians caused by the failure to diagnose whole-body MRI in the prior art.
In order to achieve the above purpose, the technical scheme of the invention is realized as follows:
In one aspect, the invention provides an intelligent whole-body tumor MRI evaluation system, which comprises an image information management module, an image recognition module, an image quality judgment module, a variation judgment module, an anatomy segmentation module, a target focus recognition module and a structural report module, wherein the image information management module is connected with the image recognition module and is used for transmitting a DICOM image of a patient to the image recognition module through a DICOM protocol when the patient shoots a whole-body magnetic resonance examination item; the image identification module is connected with the image information management module, the image quality judgment module and the structural report module and is used for identifying DICOM images matched with the whole-body magnetic resonance examination item, extracting all part information based on DICOM image header file information, extracting DICOM image sequences required by each part based on a first preset rule, defining the extracted DICOM image sequences as first images, sending the first images to the image quality judgment module, and sending the part information, the sequence types of the first images corresponding to the parts and the first images to the structural report module; when the DICOM image is not matched with the whole-body magnetic resonance examination item, stopping the diagnosis process and sending a first prompt message to the structural report module; the image quality judging module is respectively connected with the image identifying module, the variation judging module and the structural reporting module and is used for carrying out quality analysis on the first image based on preset conditions, respectively sending the first image meeting the preset conditions to the variation judging module and the structural reporting module, and defining the first image meeting the preset conditions as a second image; if the quality of the first image does not meet the preset condition, stopping the diagnosis flow; sending the judging result and the second prompt information to a structural report module; the variation judging module is connected with the image quality judging module, the anatomy segmentation module and the structural report module and is used for judging whether the second image has postoperative change and/or congenital development variation, if the second image has no postoperative change and/or congenital development variation, the second image is respectively sent to the anatomy segmentation module and the structural report module, if the second image has postoperative change and/or congenital development variation, the type of the postoperative change and/or congenital development variation is identified, each parameter of a postoperative residual structure is measured, and the type of the postoperative change and/or congenital development variation and all parameters are sent to the structural report module; the anatomical segmentation module is respectively connected with the variation judgment module, the target focus recognition module and the structural report module, and is used for segmenting all target organs and all target tissues on the second image based on a second preset rule, setting anatomical coordinates of each target organ and each target tissue, outputting first diagnosis data, setting anatomical labels for each anatomical coordinate, outputting a region of each anatomical label, namely a third image, and respectively sending the first diagnosis data and the third image to the target focus recognition module and the structural report module; the target focus identification module is respectively connected with the anatomy segmentation module and the structural report module and is used for segmenting all focuses on a third image based on a third preset rule and first diagnosis data, setting focus coordinates, namely second diagnosis data, setting focus labels for each focus coordinate, outputting the area of each focus label, namely fourth image, comparing the second diagnosis data with the first diagnosis data, positioning and measuring each focus, and transmitting the positioning result, focus measurement value, key image, second diagnosis data and fourth image of each focus to the structural report module; the structural report module is respectively connected with the image recognition module, the image quality judgment module, the variation judgment module, the anatomical segmentation module and the target focus recognition module and is used for automatically generating diagnosis impressions for doctors to check based on built-in rules of focus positioning results and focus measurement values; and stores all data and all images received.
Preferably, the anatomical segmentation module further comprises a label judging unit, configured to judge whether the anatomical label is compliant based on the first diagnostic data, where the rule of judgment is: judging whether the radial line volume of the maximum connected domain of each anatomical label is within a preset threshold value, judging whether the spatial relationship of the adjacent anatomical labels is correct, judging the shapes of different anatomical labels, and transmitting the judging result to a structural report module; if the anatomical label is compliant, the first diagnostic data is sent to the target lesion recognition module, and if the anatomical label is non-compliant, the third hint information, the type of non-compliant, and the measured value of the anatomical label are sent to the structured report module.
Preferably, the target lesion recognition module further comprises a determining unit, configured to determine a relative position of the lesion and the target organ or the target tissue based on a positioning result of the lesion, and output related data, that is: the focus is inside the target organ or target tissue, the focus invades the target organ or target tissue, and the focus is outside the target organ or target tissue.
Preferably, the target lesion recognition module further comprises a key image generation unit, configured to compare sizes of all lesions in each target organ or each target tissue based on the lesion measurement values, generate a key image of the lesions conforming to the fourth preset rule, and send the key image to the structural report module.
On the other hand, the invention also provides an MRI intelligent evaluation method for the whole-body tumor, which comprises the following steps: when the patient finishes shooting the whole-body magnetic resonance examination project, the image information management module transmits the DICOM image of the patient to the image recognition module through a DICOM protocol; the image recognition module recognizes DICOM images matched with whole-body magnetic resonance examination items, extracts all part information based on DICOM image header file information, extracts DICOM image sequences required by each part based on a first preset rule, defines the extracted DICOM image sequences as first images, sends the first images to the image quality judgment module, and sends the part information, the sequence types of the first images corresponding to the parts and the first images to the structural report module; when the DICOM image is not matched with the whole-body magnetic resonance examination item, stopping the diagnosis process and sending a first prompt message to the structural report module; the image quality judging module analyzes the quality of the first image based on preset conditions, and respectively sends the first image meeting the preset conditions to the variation judging module and the structural reporting module, and the first image meeting the preset conditions is defined as a second image; if the quality of the first image does not meet the preset condition, stopping the diagnosis flow; sending the judging result and the second prompt information to a structural report module; the variation judging module judges whether the second image has postoperative change and/or congenital development variation, if the second image has no postoperative change and/or congenital development variation, the second image is respectively sent to the anatomy segmentation module and the structural report module, if the second image has postoperative change and/or congenital development variation, the type of the postoperative change and/or congenital development variation is identified, various parameters of a residual structure after the operation are measured, and the type of the postoperative change and/or congenital development variation and all the parameters are sent to the structural report module; the anatomical segmentation module segments all target organs and all target tissues on a second image based on a second preset rule, sets anatomical coordinates of each target organ and each target tissue, outputs first diagnosis data, sets anatomical labels for each anatomical coordinate, outputs a region of each anatomical label, namely a third image, and respectively sends the first diagnosis data and the third image to the target focus recognition module and the structured report module; the target focus recognition module is used for dividing all focuses on a third image based on a third preset rule and first diagnosis data, setting focus coordinates, namely second diagnosis data, for each focus, setting focus labels for each focus coordinate, outputting a region of each focus label, namely fourth image, comparing the second diagnosis data with the first diagnosis data, positioning and measuring each focus, and transmitting a positioning result, focus measurement values, a key image, second diagnosis data and fourth image of each focus to the structural report module; the structural report module automatically generates diagnosis impressions for doctors to check the focus positioning results and focus measurement values based on built-in rules; and stores all data and all images received.
Preferably, the method further comprises: the label judging unit in the anatomical segmentation module judges whether the anatomical label is compliant based on the first diagnosis data, and the judging rule is as follows: judging whether the radial line volume of the maximum connected domain of each anatomical label is within a preset threshold value, judging whether the spatial relationship of the adjacent anatomical labels is correct, judging the shapes of different anatomical labels, and transmitting the judging result to a structural report module; if the anatomical label is compliant, the first diagnostic data is sent to the target lesion recognition module, and if the anatomical label is non-compliant, the third hint information, the type of non-compliant, and the measured value of the anatomical label are sent to the structured report module.
Preferably, the method further comprises: the judging unit in the target focus identifying module judges the relative position of the focus and the target organ or the target tissue based on the positioning result of the focus and outputs related data, namely: the focus is inside the target organ or target tissue, the focus invades the target organ or target tissue, and the focus is outside the target organ or target tissue.
Preferably, the method further comprises: the key image generating unit in the target focus identifying module compares the sizes of all focuses in each target organ or each target tissue based on focus measured values, generates key images of focuses conforming to a fourth preset rule, and sends the key images to the structural reporting module.
The invention has the technical effects that:
1. Because the invention sets up the image recognition module, image quality judging module, variation judging module, dissecting and dividing module, goal focus recognition module and structuring report module, when the patient shoots the whole body magnetic resonance (WB-MRI) examination project, the image recognition module will discern DIOCOM image-related DICOM image sequence, the image quality judging module will analyze, judge the quality of the required DICOM image sequence, will be because of poor quality DICOM image recognition caused by artifact, etc., in order to prevent influencing the subsequent diagnosis, the variation judging module will analyze the qualified DICOM image sequence after operation, if will not reject the anatomical change and congenital abnormal DICOM image caused by operation in advance, will cause a large amount of erroneous judgement of focus analysis model of the subsequent diagnosis flow, have influenced the diagnosis precision, the dissecting module will divide the target organ and or target tissue to the DICOM image which has changed after no operation, the goal focus recognition module will divide, position the focus, the structured report module will integrate all processed data and images and store, output the diagnosis impression for reference; the system correlates the evaluation part and the part evaluation method with a plurality of AI diagnostic models, and embeds the related evaluation method in the structural report module, thereby greatly reducing the recording strength, improving the diagnosis efficiency of doctors and reducing the examination cost of patients; the original impractical evaluation scheme can be changed into a floor type; with the use of an accurate whole-body heterogeneity evaluation system, the selection of a treatment scheme is changed, high-level research evidence is accumulated, and an informatization tool and an intelligent technology are received in the future to integrate image information and clinical information together, so that better clinical decisions can be obtained, and the clinical value of image services is improved;
2. Because the invention sets the label judging unit, can judge whether the anatomical label is compliant based on the first diagnostic data, if the anatomical label is compliant, send the first diagnostic data to the focus identification module of goal, if the anatomical label is not compliant, send the third prompt message to the structural report module, have prevented the diagnosis inaccuracy caused by the unqualified anatomical coordinates, meanwhile, have prompt message to send, in order to let the manual intervention, process in time, perfect AI diagnostic model, make the whole diagnostic procedure more perfect, more systematic;
3. because the invention is provided with the judging unit, the relative position of the focus and the target organ or the target tissue can be judged based on the positioning result of the focus, and the related data is output, namely: the focus is in the target organ or the target tissue, the focus invades the target organ or the target tissue, and the focus is outside the target organ or the target tissue, and the focus is synchronously fed back to the corresponding interface of the structural report module, thereby being beneficial to the diagnosis of clinicians;
4. Because the invention is provided with the key image generating unit, the size of all focuses in each target organ or each target tissue can be compared based on the focus measured value, the focus conforming to the fourth preset rule (such as the largest 3 focuses of each part) is generated into the key image, and the key image is sent to the structural report module, so that the diagnosis efficiency of a clinician is improved, and the structural report interface is more visual.
Detailed Description
The invention will be described in detail below with reference to the drawings in combination with embodiments.
Example 1
Fig. 1 shows a schematic structural diagram of a whole-body tumor MRI intelligent assessment system according to a first embodiment of the present invention; as shown in fig. 1, the system includes: an image information management module 10, an image recognition module 20, an image quality determination module 30, a variance determination module 40, an anatomical segmentation module 50, a target lesion recognition module 60, and a structured reporting module 70, wherein,
The image information management module 10 is connected with the image recognition module 20 and is used for transmitting the DICOM image of the patient to the image recognition module 20 through a DICOM protocol when the patient shoots the whole body magnetic resonance examination item;
wherein, the image information management module is RIS (Radiology Information System) systems.
The image recognition module 20 is connected with the image information management module 10, the image quality judging module 30 and the structural report module 70, and is used for recognizing a DICOM image matched with a whole body magnetic resonance examination item, extracting all part information based on DICOM image header file information, extracting a DICOM image sequence required by each part based on a first preset rule, defining the extracted DICOM image sequence as a first image, sending the first image to the image quality judging module 30, and sending part information, the sequence type of the first image corresponding to the part and the first image to the structural report module 70; when the DICOM image does not match the whole body magnetic resonance examination item, the diagnostic procedure is aborted and a first prompt is sent to the structured reporting module 70;
The site refers to a target organ or target tissue, such as lymph node, organ, prostate, bone, etc., and the site information and the sequence type corresponding to the site are returned to the corresponding control of the "technical evaluation (scanning sequence)" of the structured report interface according to the requirement.
The DICOM image sequences may be dwi_high, dwi_low, ADC, t1wi_in, t1wi_opp, t2wi_fs, T2WI, etc., and all DICOM image sequences required for each location are identified for subsequent AI model diagnosis. When the DICOM image is not matched with the whole-body magnetic resonance examination project, the diagnosis process is stopped, first prompt information is sent to the structural report module, related responsible personnel are prompted to process in time, and the first image and the first prompt information are stored in a database of the structural report module.
The image quality judging module 30 is respectively connected with the image identifying module 20, the variation judging module 40 and the structured reporting module 70, and is used for carrying out quality analysis on the first image based on preset conditions, respectively sending the first image meeting the preset conditions to the variation judging module 40 and the structured reporting module 70, and defining the first image meeting the preset conditions as a second image; if the quality of the first image does not meet the preset condition, stopping the diagnosis flow; and sends the judgment result and the second prompt message to the structural report module 70;
And recognizing DICOM images with unqualified quality, and avoiding the conditions of image diagnosis, such as low image signal-to-noise ratio, magnetic sensitivity artifact, respiratory artifact, motion artifact, complement of scanning range, incomplete image sequence and the like of DWI. If the DICOM image is unqualified, the diagnosis process is stopped, a second prompt message is sent to the structural report module, a judging result is sent to a corresponding control of 'technical evaluation (image quality)' of the structural report interface, relevant responsible personnel are prompted to timely process, and the second image and the second prompt message are stored in a database of the structural report module.
The variation judging module 40 is connected to the image quality judging module 30, the anatomical segmentation module 50 and the structural report module 70, and is configured to judge whether the second image has a post-operation change and/or a congenital development variation, if not, send the second image to the anatomical segmentation module 50 and the structural report module 70, and if so, identify the type of the post-operation change and/or the congenital development variation, measure various parameters of the residual structure after operation, and send the type of the post-operation change and/or the congenital development variation and all parameters to the structural report module 70;
Determining if there is a post-treatment change that alters the anatomy, such as: TURP, radical prostatectomy, other pelvic procedures, etc., are fed back into the overall assessment of the structured report interface.
The types of postoperative changes and/or congenital developmental variations are TURP, radical prostatectomy, other pelvic surgery, etc., and various parameters measured, such as the volume of the residual structure after the operation, radial lines, etc., are judged to have the postoperative changes, and the postoperative changes or the congenital developmental variations are sent to the corresponding controls of the overall evaluation (postoperative changes, congenital developmental variations) of the structured report interface.
The anatomy segmentation module 50 is respectively connected with the mutation judgment module 40, the target focus recognition module 60 and the structured report module 70, and is configured to segment all target organs and all target tissues on the second image based on a second preset rule, set anatomical coordinates of each target organ and each target tissue, output first diagnosis data, set anatomical labels for each anatomical coordinate, output a region of each anatomical label, namely a third image, and send the first diagnosis data and the third image to the target focus recognition module 60 and the structured report module 70 respectively;
for example, for whole-body magnetic resonance examination of prostate cancer metastasis, the segmentation of target organs and target tissues is as follows:
Dividing the pelvic cavity soft tissue structures of the prostate, seminal vesicle gland, rectum, bladder, obturator internus muscle, levator ani muscle and the like;
Dividing soft tissue structures of abdomen, chest and head;
Dividing the whole body lymph node area including pelvic lymph node, retroperitoneal lymph node, other lymph nodes;
Pelvic bone structures such as lower lumbar vertebra, ilium, sacrum, ischium, pubis, acetabulum, femoral neck, femoral head, femoral neck and the like are divided, and bone structures such as cranium, cervical vertebra, thoracic vertebra, lumbar vertebra, rib, thoracic rib and the like are divided.
RECIST 1.1 efficacy evaluation criteria were used for three tissues/organs of lymph node, organ, and prostate, and MET-RADS-P custom criteria (progression, stabilization, response) were used for skeletal treatment response.
Comprehensive RAC evaluation rules: referring to Prostate Cancer Working Group, the RAC score of PCWG3 defines a standard table, which is not described in detail. The present systematic approach embeds these clear rules into the structured report, allowing it to be automated to generate the RAC score.
The target focus identifying module 60 is respectively connected with the anatomy segmentation module 50 and the structural report module 70, and is configured to segment all focuses on the third image based on a third preset rule and the first diagnostic data, set focus coordinates, namely second diagnostic data, set focus labels for each focus coordinate, output a region, namely fourth image, of each focus label, compare the second diagnostic data with the first diagnostic data, locate and measure each focus, and send a locating result, focus measurement value, a key image, second diagnostic data and fourth image of each focus to the structural report module 70;
For bone lesions, lymph node lesions, soft tissue lesions, prostate lesions, segmentation is performed using different image sequences with AI models; this technique, although highly complex, is relatively well established and will not be described in detail. The result of the segmentation is the coordinates of the anatomical label of the lesion and the measured value information of the lesion.
The structural report module 70 is respectively connected with the image recognition module 20, the image quality judgment module 30, the variation judgment module 40, the anatomical segmentation module 50 and the target focus recognition module 60, and is used for automatically generating diagnosis impressions of focus positioning results and focus measurement values based on built-in rules for a doctor to check; and stores all data and all images received.
Wherein the localization results and lesion measurements are returned to the "diagnostic impression" of the structured report.
Fig. 2 shows a schematic structural report interface of a whole-body MRI for prostate cancer metastasis in a schematic structural diagram of a whole-body MRI intelligent evaluation system according to a first embodiment of the present invention; as shown in fig. 2, the technical assessment includes image sequences and image quality, structured reports and automatic association of individual AI modules, and the overall assessment includes a list of corresponding parameters of the target organ and target tissue foci, including primary organ (prostate), liver, lung, other soft tissue areas, pelvic lymph nodes, retroperitoneal lymph nodes, other lymph nodes, bones, skull, cervical vertebrae, thoracic vertebrae, sacral vertebrae, pelvis, thorax, limbs.
The whole body MRI examination (WB-MRI) uses basic T1, T2 and DWI sequences (including ADC images; the DWI sequences can display not only tumor in glands but also lymph node metastasis and bone metastasis), and quantitative detection of bones and lymph node metastasis can be completed within 30 minutes. If the detection of the software tissue and the internal organs is increased, only 45-50 minutes is required. The method has low cost, no additional damage and no clinical effect inferior to PET-CT, so that a standard WB-MRI evaluation method has more clinical value.
The evaluation of WB-MRI is plagued in clinical use by its numerous evaluation sites, and the evaluation rules for each site are related to the disease type. Without standardized, intelligent tools, proper assessment is essentially impossible with the memory capacity of the physician alone. Taking WB-MRI prostate as an example, the range of scan and evaluation required includes: bone, lymph nodes, soft tissue and viscera of head and neck trunk, and tumor of peripheral pelvic part of prostate.
For tumors of bones, lymph, soft tissues, organs, methods for automatic feature extraction using specific AI, structured reporting methods for single lesion characterization, and clinical evaluation of lesion changes over time have all been mature. The present systematic method organizes prostate cancer systemic metastasis assessment according to their logic: the method comprises the steps of evaluating the focus sequentially, and integrating the characteristics of the post-treatment/AI automatic extraction into a report if the post-treatment/AI automatic extraction exists; finally, scores of 1-5 points were automatically given according to the RAC of WB-MRI-P (response assessment category, type of response assessment), providing a clinical reference.
The target lesion segmentation information and the target lesion localization information are sent to a structured reporting system for whole-body MRI prostate assessment. The structured report system embeds the RECIST evaluation method, the bone reaction evaluation method and the RAC scoring method in the structured report system through conventional programs, and then automatically obtains an evaluation conclusion according to data (or manually input values) transmitted by the AI diagnostic modules.
The system can be used for evaluating the systemic metastasis of WB-MRI-P prostate cancer, and can be used for evaluating tumors such as breast cancer, lymphoma, hematopathy and the like through configuration rules.
The embodiment of the invention is provided with an image recognition module, an image quality judgment module, a variation judgment module, an anatomical segmentation module, a target focus recognition module and a structured report module, when a patient shoots a whole body magnetic resonance (WB-MRI) examination project, the image recognition module recognizes DIOCOM image-related DICOM image sequences, the image quality judgment module analyzes and judges the quality of the required DICOM image sequences, poor-quality DICOM images caused by artifacts and the like are recognized so as to prevent the influence on subsequent diagnosis, the variation judgment module performs postoperative change analysis on the qualified DICOM image sequences, if anatomical changes caused by operations and congenital malformed DICOM images are not removed in advance, a great deal of misjudgment of a focus analysis model of the subsequent diagnosis flow is caused, diagnosis accuracy is influenced, the anatomical segmentation module segments and positions focuses of target organs and or target tissues of the DICOM images which are not changed after operation, and the structural report module integrates all processed data and images and stores and outputs diagnosis impressions for reference; the system correlates the evaluation part and the part evaluation method with a plurality of AI diagnostic models, and embeds the related evaluation method in the structural report module, thereby greatly reducing the recording strength, improving the diagnosis efficiency of doctors and reducing the examination cost of patients; the original impractical evaluation scheme can be changed into a floor type; with the use of an accurate whole-body heterogeneity evaluation system, the selection of a treatment scheme is changed, high-level research evidence is accumulated, and an informatization tool and an intelligent technology are received in the future to integrate image information and clinical information together, so that better clinical decisions can be obtained, and the clinical value of image services is improved.
Example two
Fig. 3 is a schematic structural diagram of a whole-body tumor MRI intelligent assessment system according to a second embodiment of the present invention, as shown in fig. 3, the anatomical segmentation module 50 further includes a label judging unit 502, configured to judge whether an anatomical label is compliant based on the first diagnostic data, where a rule of judgment is: judging whether the radial line volume of the maximum connected domain of each anatomical label is within a preset threshold value, judging whether the spatial relationship of the adjacent anatomical labels is correct, judging the shapes of different anatomical labels, and transmitting the judging result to the structural report module 70; if the anatomical label is compliant, the first diagnostic data is sent to the target lesion recognition module 60, and if the anatomical label is non-compliant, the third hint information, the type of non-compliance, and the measurement of the anatomical label are sent to the structured reporting module 70.
For example, it is determined whether the radial volume of the largest connected region of each anatomical label is within a preset threshold, such as 10% -90% of the preset threshold, if the first diagnostic data outside the range is not compliant.
The embodiment of the invention is provided with the label judging unit, can judge whether the anatomical label is in compliance based on the first diagnosis data, if the anatomical label is in compliance, the first diagnosis data is sent to the target focus segmentation module, and if the anatomical label is in non-compliance, the third prompt information is sent to the structural report module, so that the inaccuracy of diagnosis caused by unqualified anatomical coordinates is avoided, and meanwhile, the prompt information is sent, so that manual intervention and processing are facilitated in time, an AI diagnosis model is perfected, the whole diagnosis flow is more perfect, and the system is realized.
Example III
Fig. 4 is a schematic structural diagram of a whole-body tumor MRI intelligent evaluation system according to a third embodiment of the present invention, as shown in fig. 4, the target lesion recognition module 60 further includes a determining unit 602, configured to determine a relative position of a lesion and a target organ or a target tissue based on a positioning result of the lesion, and output related data, that is: the focus is inside the target organ or target tissue, the focus invades the target organ or target tissue, and the focus is outside the target organ or target tissue.
The embodiment of the invention is provided with the judging unit which can judge the relative position of the focus and the target organ or the target tissue based on the positioning result of the focus and output related data, namely: the focus is in the target organ or the target tissue, the focus invades the target organ or the target tissue, and the focus is outside the target organ or the target tissue, and the focus is synchronously fed back to the corresponding interface of the structural report module, thereby being beneficial to the diagnosis of clinicians.
Example IV
Fig. 5 is a schematic structural diagram of a whole-body tumor MRI intelligent assessment system according to a fourth embodiment of the present invention, and as shown in fig. 5, the target lesion recognition module 60 further includes a key image generating unit 604 for comparing sizes of all lesions in each target organ or each target tissue based on the lesion measurement values, generating a key image of the lesions conforming to a fourth preset rule, and transmitting the key image to the structural report module 70.
For example, a key image is generated for the largest 3 lesions per site.
The embodiment of the invention is provided with the key image generation unit, can compare the sizes of all focuses in each target organ or each target tissue based on the focus measured value, generate key images of the focuses conforming to a fourth preset rule (such as the largest 3 focuses of each part), and send the key images to the structural report module, thereby improving the diagnosis efficiency of clinicians and enabling the structural report interface to be more visual.
Example five
FIG. 6 shows a flowchart of a whole-body tumor MRI intelligent assessment method according to a fifth embodiment of the present invention; as shown in fig. 6, the method comprises the steps of:
step S501, when the patient finishes shooting the whole body magnetic resonance examination project, the image information management module transmits the DICOM image of the patient to the image recognition module through a DICOM protocol;
wherein the image information management module is RIS (Radiology Information System) systems;
Step S502, an image recognition module recognizes DICOM images matched with whole-body magnetic resonance examination items, extracts all part information based on DICOM image header file information, extracts DICOM image sequences required by each part based on a first preset rule, defines the extracted DICOM image sequences as first images, sends the first images to an image quality judgment module, and sends part information, the sequence type of the first images corresponding to the parts and the first images to a structural report module; when the DICOM image is not matched with the whole-body magnetic resonance examination item, stopping the diagnosis process and sending a first prompt message to the structural report module;
The site refers to a target organ or target tissue, such as lymph node, organ, prostate, bone, etc., and the site information and the sequence type corresponding to the site are returned to the corresponding control of the "technical evaluation (scanning sequence)" of the structured report interface according to the requirement.
The DICOM image sequences may be dwi_high, dwi_low, ADC, t1wi_in, t1wi_opp, t2wi_fs, T2WI, etc., and all DICOM image sequences required for each location are identified for subsequent AI model diagnosis. When the DICOM image is not matched with the whole-body magnetic resonance examination project, the diagnosis process is stopped, first prompt information is sent to the structural report module, related responsible personnel are prompted to process in time, and the first image and the first prompt information are stored in a database of the structural report module.
Step S503, the image quality judging module analyzes the quality of the first image based on the preset condition, and sends the first image meeting the preset condition to the variation judging module and the structured report module respectively, and the first image meeting the preset condition is defined as a second image; if the quality of the first image does not meet the preset condition, stopping the diagnosis flow; sending the judging result and the second prompt information to a structural report module;
And recognizing DICOM images with unqualified quality, and avoiding the conditions of image diagnosis, such as low image signal-to-noise ratio, magnetic sensitivity artifact, respiratory artifact, motion artifact, complement of scanning range, incomplete image sequence and the like of DWI. If the DICOM image is unqualified, the diagnosis process is stopped, a second prompt message is sent to the structural report module, a judging result is sent to a corresponding control of 'technical evaluation (image quality)' of the structural report interface, relevant responsible personnel are prompted to timely process, and the second image and the second prompt message are stored in a database of the structural report module.
Step S504, the variation judging module judges whether the second image has postoperative change and/or congenital development variation, if not, the second image is respectively sent to the anatomy segmentation module and the structural report module, if so, the type of the postoperative change and/or the congenital development variation is identified, various parameters of the postoperative residual structure are measured, and the type of the postoperative change and/or the congenital development variation and all the parameters are sent to the structural report module;
Determining if there is a post-treatment change that alters the anatomy, such as: TURP, radical prostatectomy, other pelvic procedures, etc., are fed back into the overall assessment of the structured report interface.
The types of postoperative changes and/or congenital developmental variations are TURP, radical prostatectomy, other pelvic surgery, etc., and various parameters measured, such as the volume of the residual structure after the operation, radial lines, etc., are judged to have the postoperative changes, and the postoperative changes or the congenital developmental variations are sent to the corresponding controls of the overall evaluation (postoperative changes, congenital developmental variations) of the structured report interface.
Step S505, the anatomy segmentation module segments all target organs and all target tissues on a second image based on a second preset rule, sets the anatomical coordinates of each target organ and each target tissue, outputs first diagnosis data, sets anatomical labels for each anatomical coordinate, outputs a region of each anatomical label, namely a third image, and respectively sends the first diagnosis data and the third image to the target focus recognition module and the structured report module;
for example, for whole-body magnetic resonance examination of prostate cancer metastasis, the segmentation of target organs and target tissues is as follows:
Dividing the pelvic cavity soft tissue structures of the prostate, seminal vesicle gland, rectum, bladder, obturator internus muscle, levator ani muscle and the like;
Dividing soft tissue structures of abdomen, chest and head;
Dividing the whole body lymph node area including pelvic lymph node, retroperitoneal lymph node, other lymph nodes;
Pelvic bone structures such as lower lumbar vertebra, ilium, sacrum, ischium, pubis, acetabulum, femoral neck, femoral head, femoral neck and the like are divided, and bone structures such as cranium, cervical vertebra, thoracic vertebra, lumbar vertebra, rib, thoracic rib and the like are divided.
RECIST 1.1 efficacy evaluation criteria were used for three tissues/organs of lymph node, organ, and prostate, and MET-RADS-P custom criteria (progression, stabilization, response) were used for skeletal treatment response.
Comprehensive RAC evaluation rules: referring to Prostate Cancer Working Group, the RAC score of PCWG3 defines a standard table, which is not described in detail. The present systematic approach embeds these clear rules into the structured report, allowing it to be automated to generate the RAC score.
Step S506, the target focus recognition module segments all focuses on a third image based on a third preset rule and first diagnosis data, sets focus coordinates, namely second diagnosis data, sets focus labels for each focus coordinate, outputs a region of each focus label, namely fourth image, compares the second diagnosis data with the first diagnosis data, positions and measures each focus, and sends a positioning result, focus measurement value, a key image, second diagnosis data and fourth image of each focus to the structural report module;
For bone lesions, lymph node lesions, soft tissue lesions, prostate lesions, segmentation is performed using different image sequences with AI models; this technique, although highly complex, is relatively well established and is not described in detail. The result of the segmentation is the coordinates of the anatomical label of the lesion and the measured value information of the lesion.
Step S507, the structural report module automatically generates diagnosis impressions for doctors to check based on the built-in rules and the focus positioning results and focus measurement values; and stores all data and all images received.
Wherein the localization results and lesion measurements are returned to the "diagnostic impression" of the structured report.
FIG. 7 is a schematic diagram showing a structured report interface of a whole-body MRI (magnetic resonance imaging) examination for prostate cancer metastasis in a method for intelligently evaluating whole-body tumors according to a fifth embodiment of the present invention; as shown in fig. 7, the technical assessment includes image sequences and image quality, structured reports and automatic association of individual AI modules, and the overall assessment includes a list of corresponding parameters of the target organ and target tissue foci, including primary organ (prostate), liver, lung, other soft tissue areas, pelvic lymph nodes, retroperitoneal lymph nodes, other lymph nodes, bone, skull, cervical, thoracic, sacral, pelvic, thoracic, limb.
The whole body MRI examination (WB-MRI) uses basic T1, T2 and DWI sequences (including ADC images; the DWI sequences can display not only tumor in glands but also lymph node metastasis and bone metastasis), and quantitative detection of bones and lymph node metastasis can be completed within 30 minutes. If the detection of the software tissue and the internal organs is increased, only 45-50 minutes is required. The method has low cost, no additional damage and no clinical effect inferior to PET-CT, so that a standard WB-MRI evaluation method has more clinical value.
The evaluation of WB-MRI is plagued in clinical use by its numerous evaluation sites, and the evaluation rules for each site are related to the disease type. Without standardized, intelligent tools, proper assessment is essentially impossible with the memory capacity of the physician alone. Taking WB-MRI prostate as an example, the range of scan and evaluation required includes: bone, lymph nodes, soft tissue and viscera of head and neck trunk, and tumor of peripheral pelvic part of prostate.
For tumors of bones, lymph, soft tissues, organs, methods for automatic feature extraction using specific AI, structured reporting methods for single lesion characterization, and clinical evaluation of lesion changes over time have all been mature. The present systematic method organizes prostate cancer systemic metastasis assessment according to their logic: the method comprises the steps of evaluating the focus sequentially, and integrating the characteristics of the post-treatment/AI automatic extraction into a report if the post-treatment/AI automatic extraction exists; finally, scores of 1-5 points were automatically given according to the RAC of WB-MRI-P (response assessment category, type of response assessment), providing a clinical reference.
The target lesion segmentation information and the target lesion localization information are sent to a structured reporting system for whole-body MRI prostate assessment. The structured report system embeds the RECIST evaluation method, the bone reaction evaluation method and the RAC scoring method in the structured report system through conventional programs, and then automatically obtains an evaluation conclusion according to data (or manually input values) transmitted by the AI diagnostic modules.
The system can be used for evaluating the systemic metastasis of WB-MRI-P prostate cancer, and can be used for evaluating tumors such as breast cancer, lymphoma, hematopathy and the like through configuration rules.
Wherein the method further comprises: the label judging unit in the anatomical segmentation module judges whether the anatomical label is compliant based on the first diagnosis data, and the judging rule is as follows: judging whether the radial line volume of the maximum connected domain of each anatomical label is within a preset threshold value, judging whether the spatial relationship of the adjacent anatomical labels is correct, judging the shapes of different anatomical labels, and transmitting the judging result to a structural report module; if the anatomical label is compliant, the first diagnostic data is sent to the target lesion recognition module, and if the anatomical label is non-compliant, the third hint information, the type of non-compliant, and the measured value of the anatomical label are sent to the structured report module.
For example, it is determined whether the radial volume of the largest connected region of each anatomical label is within a preset threshold, such as 10% -90% of the preset threshold, if the first diagnostic data outside the range is not compliant.
Wherein the method further comprises: the judging unit in the target focus identifying module judges the relative position of the focus and the target organ or the target tissue based on the positioning result of the focus and outputs related data, namely: the focus is inside the target organ or target tissue, the focus invades the target organ or target tissue, and the focus is outside the target organ or target tissue.
Wherein the method further comprises: the key image generating unit in the target focus identifying module compares the sizes of all focuses in each target organ or each target tissue based on focus measured values, generates key images of focuses conforming to a fourth preset rule, and sends the key images to the structural reporting module.
For example, a key image is generated for the largest 3 lesions per site.
The method comprises the steps that when a patient shoots a whole body magnetic resonance (WB-MRI) examination item, the image recognition module recognizes DIOCOM image related DICOM image sequences, the image quality judgment module analyzes and judges the quality of the required DICOM image sequences, poor-quality DICOM images caused by artifacts and the like are recognized so as to prevent the influence on subsequent diagnosis, the variation judgment module performs postoperative change analysis on the qualified DICOM image sequences, if anatomical changes caused by operation and congenital malformed DICOM images are not removed in advance, a large number of misjudgment of a focus analysis model of the subsequent diagnosis flow is caused, diagnosis accuracy is influenced, the anatomical segmentation module segments target organs and/or target tissues on the DICOM images which are not changed after operation, the target focus recognition module segments and positions focuses, and the structural report module integrates all processed data and images and stores and outputs diagnosis impressions for reference; the system correlates the evaluation part and the part evaluation method with a plurality of AI diagnostic models, and embeds the related evaluation method in the structural report module, thereby greatly reducing the recording strength, improving the diagnosis efficiency of doctors and reducing the examination cost of patients; the original impractical evaluation scheme can be changed into a floor type; with the use of an accurate whole-body heterogeneity evaluation system, the selection of a treatment scheme is changed, high-level research evidence is accumulated, and an informatization tool and an intelligent technology are received in the future to integrate image information and clinical information together, so that better clinical decisions can be obtained, and the clinical value of image services is improved; because the label judging unit in the embodiment of the invention can judge whether the anatomical label is in compliance based on the first diagnosis data, if the anatomical label is in compliance, the first diagnosis data is sent to the target focus segmentation module, and if the anatomical label is in non-compliance, the third prompt information is sent to the structural report module, thereby avoiding inaccurate diagnosis caused by unqualified anatomical coordinates, and meanwhile, the prompt information is sent, so that manual intervention and processing can be timely carried out, an AI diagnosis model is perfected, the whole diagnosis flow is more perfect and systematic; because the judging unit in the embodiment of the invention can judge the relative position of the focus and the target organ or the target tissue based on the positioning result of the focus, and output related data, namely: the focus is in the target organ or the target tissue, the focus invades the target organ or the target tissue, and the focus is outside the target organ or the target tissue, and the focus is synchronously fed back to the corresponding interface of the structural report module, thereby being beneficial to the diagnosis of clinicians; because the key image generating unit in the invention can compare the sizes of all focuses in each target organ or each target tissue based on the focus measured value, generate key images of focuses conforming to a fourth preset rule (such as the largest 3 focuses of each position), and send the key images to the structural report module, the diagnosis efficiency of a clinician is improved, and the structural report interface is more visual.
Example six
Fig. 8 shows a specific process flow diagram of a whole-body tumor MRI intelligent assessment method according to a sixth embodiment of the present invention, as shown in fig. 8, the method comprising the steps of:
In step S601, the image recognition module recognizes that the DICOM image matches the examination item (whole body magnetic resonance imaging)? If not, the first prompt message is sent to the structural report module, if yes, the step S602 is executed;
step S602, an image recognition module extracts a required image sequence, sends a first image, a sequence type and position information to a structural report module, and sends the first image to an image quality judging module;
Step S603, the image quality is in accordance with the preset condition? If not, sending the judging result and the second prompt information to a structural report module; if yes, go to step S604;
step S604, is there a post-operative change and/or an congenital variation? If yes, the post-operation change and/or the congenital variation and the post-operation residual structure measurement parameters are sent to a structural report module, and if not, step S605 is executed;
Step S605, the anatomy segmentation module segments the target organ and the target tissue, sends the first diagnosis data and the third image to the result report module and performs step S606;
Step S606, if not, the third prompt information, the type of non-compliance and the measured value of the anatomical label are sent to the structural report module, if yes, step S607 is executed;
Step S607, the target focus identifying module segments focus, positions and measures focus, and sends focus positioning result, focus measured value, key image, second diagnosis data and fourth image to the structural report module;
Step S608, the structural report module automatically generates diagnosis impressions for doctors to check based on the built-in rules and the focus positioning results and focus measured values; and stores all data and all images received.
From the above description, it can be seen that the above embodiments of the present invention achieve the following technical effects: because the embodiment of the invention is provided with the image recognition module, the image quality judgment module, the variation judgment module, the anatomical segmentation module, the target focus recognition module and the structural report module, when a patient shoots a whole body magnetic resonance (WB-MRI) examination project, the image recognition module recognizes DIOCOM image related DICOM image sequences, the image quality judgment module analyzes and judges the quality of the required DICOM image sequences, the poor quality DICOM images caused by artifacts and the like are recognized to prevent the influence on subsequent diagnosis, the variation judgment module analyzes the qualified DICOM image sequences after operation, if anatomical changes caused by operation and the prior abnormal DICOM images are not removed in advance, a great deal of misjudgment of a focus analysis model of the subsequent diagnosis flow is caused, the diagnosis precision is influenced, the anatomical segmentation module segments target organs and/or target tissues on the DICOM images which are not changed after operation, the target focus recognition module segments and positions the images, and the structural report module integrates all processed data and images and stores the images and outputs diagnosis impressions for reference by doctors; the system correlates the evaluation part and the part evaluation method with a plurality of AI diagnostic models, and embeds the related evaluation method in the structural report module, thereby greatly reducing the recording strength, improving the diagnosis efficiency of doctors and reducing the examination cost of patients; the original impractical evaluation scheme can be changed into a floor type; with the use of an accurate whole-body heterogeneity evaluation system, the selection of a treatment scheme is changed, high-level research evidence is accumulated, and an informatization tool and an intelligent technology are received in the future to integrate image information and clinical information together, so that better clinical decisions can be obtained, and the clinical value of image services is improved; because the label judging unit is arranged in the embodiment of the invention, whether the anatomical label is in compliance or not can be judged based on the first diagnosis data, if the anatomical label is in compliance, the first diagnosis data is sent to the target focus segmentation module, and if the anatomical label is in non-compliance, the third prompt information is sent to the structural report module, so that the inaccuracy of diagnosis caused by unqualified anatomical coordinates is avoided, and meanwhile, the prompt information is sent, so that the manual intervention and processing can be timely carried out, the AI diagnosis model is perfected, the whole diagnosis flow is more perfect and systematic; because the embodiment of the invention is provided with the judging unit, the relative position of the focus and the target organ or the target tissue can be judged based on the positioning result of the focus, and the related data is output, namely: the focus is in the target organ or the target tissue, the focus invades the target organ or the target tissue, and the focus is outside the target organ or the target tissue, and the focus is synchronously fed back to the corresponding interface of the structural report module, thereby being beneficial to the diagnosis of clinicians; because the embodiment of the invention is provided with the key image generating unit, the size of all focuses in each target organ or each target tissue can be compared based on the focus measured value, the focus conforming to the fourth preset rule (such as the largest 3 focuses of each part) is generated into the key image, and the key image is sent to the structural report module, so that the diagnosis efficiency of a clinician is improved, and the structural report interface is more visual.
It will be apparent to those skilled in the art that the modules or steps of the invention described above may be implemented in a general purpose computing device, they may be concentrated on a single computing device, or distributed across a network of computing devices, or they may alternatively be implemented in program code executable by computing devices, such that they may be stored in a memory device for execution by the computing devices, or they may be separately fabricated into individual integrated circuit modules, or multiple modules or steps within them may be fabricated into a single integrated circuit module. Thus, the present invention is not limited to any specific combination of hardware and software.
The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.