A method, apparatus, and program product for predicting post-radical prostatectomy upgrading of a prostate cancer with a Gleason score of 3+3 and 3+4

By combining prostate mpMRI images and clinical data, a multi-level prediction model was used to predict Gleason score upgrades, which solved the problem of inconsistent scores after radical surgery and improved the accuracy of scoring and the scientific nature of treatment decisions.

CN120319402BActive Publication Date: 2025-12-30PEKING UNION MEDICAL COLLEGE HOSPITAL
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202510469333.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2025-04-15
Publication Date
2025-12-30
Estimated Expiration
2045-04-15

AI Technical Summary

Technical Problem

In existing technologies, the discrepancy between Gleason scores and biopsy scores after radical prostatectomy is high, making treatment decisions difficult. In particular, for patients with Gleason scores of 3+3 and 3+4, there is a lack of effective methods for accurately predicting postoperative score progression.

Method used

Based on prostate mpMRI imaging data and preoperative biopsy data, combined with the degree of extracapsular invasion and lesion location, a multi-level prediction model (such as SVM, XGBoost, convolutional neural network, etc.) is used to predict the Gleason score upgrade. The extracapsular invasion level and lesion location features are extracted by the recognition network and combined with clinical data for comprehensive prediction.

Benefits of technology

It improves the accuracy of Gleason scores after radical prostatectomy, helps doctors develop more appropriate treatment plans, reduces the risk of overtreatment or undertreatment, and improves patient outcomes.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120319402B_ABST
    Figure CN120319402B_ABST
Patent Text Reader

Abstract

The application relates to the field of intelligent medical treatment, in particular to a puncture Gleason score 3+3 and 3+4 prostate cancer radical resection postoperative score upgrading prediction method, equipment and program product. The method comprises the following steps: acquiring prostate mpMRI image data of a to-be-measured person; obtaining the position of a prostate lesion based on the image data; when the position of the prostate lesion is located at the bottom, determining postoperative upgrading on the basis of the prostate cancer Gleason score grade to obtain a prediction result. The application can perform prostate cancer radical resection postoperative Gleason score upgrading prediction based on preoperative biopsy data, and has good clinical value.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of intelligent healthcare, specifically to a method, device, program product, and computer-readable storage medium for predicting postoperative score upgrades in prostatectomy patients with Gleason scores of 3+3 and 3+4. Background Technology

[0002] Prostate cancer (PCa) is the most common cancer in men, accounting for 15% of all cancers. The Gleason score (GS) is an important risk stratification and prognostic indicator for PCa. Patients with a GS of 3+3 should undergo active surveillance (AS), intermediate-risk PCa patients with a GS of 3+4 can choose AS, while patients with a GS of 4+3 or higher are not suitable for AS and are advised to undergo radical prostatectomy (RP) combined with radiotherapy or androgen deprivation therapy. Treatment of PCa usually relies on biopsy of the GS (bGS). However, the inconsistency rate between the GS (pGS) and bGS after radical prostatectomy ranges from 20% to 60%, and accurate patient classification is crucial for making treatment decisions. However, current treatment for PCa is inadequate. Summary of the Invention

[0003] To address the aforementioned issues, establishing accurate escalation prediction models for biopsy Gleason scores of 3+3 and 3+4 is crucial for tumor risk assessment and treatment planning in PCa patients. This invention provides a method for predicting escalation of prostatectomy scores after radical prostatectomy for biopsy Gleason scores of 3+3 and 3+4, specifically including:

[0004] S1. Based on the acquisition of prostate mpMRI image data and preoperative biopsy data of the subject;

[0005] S2. Based on preoperative biopsy data, the Gleason score of the patient's prostate cancer was obtained;

[0006] S3. Based on the image data, the location of the prostate lesion is obtained; when the prostate lesion is located at the bottom, the postoperative escalation is determined based on the Gleason score of prostate cancer, and a prediction result is obtained.

[0007] The prostate cancer Gleason score is 3+3 or 3+4;

[0008] Optionally, S3 can be replaced by: obtaining the location of the prostate lesion based on the image data, and inputting the location of the prostate lesion into the first prediction model to predict whether to upgrade or not;

[0009] Optionally, the first prediction model may employ one or more of the following: SVM, XGBoost / LightGBM, convolutional neural network, or Transformer.

[0010] Optionally, the method further includes the degree of extracapsular invasion of the prostate. The degree of extracapsular invasion of the prostate is obtained based on the imaging data. When the degree of extracapsular invasion of the prostate is greater than or equal to grade II and the prostate lesion is located at the bottom, the postoperative escalation is determined based on the Gleason score of the prostate cancer to obtain a prediction result.

[0011] The method also includes the degree of extracapsular invasion of the prostate, the diameter of the prostate capsule (PCa), and the PI-RADS v2.1 score. Based on the image data, the diameter of the PCa and the PI-RADS v2.1 score are obtained. The location of the prostate lesion, the degree of extracapsular invasion of the prostate, the diameter of the PCa, and the PI-RADS v2.1 score are input into a second prediction model to predict whether the lesion will be upgraded or not.

[0012] The S3 is replaced by: inputting the image data into the target detection network to detect and obtain the location of the prostate lesion, and simultaneously inputting the image data into the recognition network to determine the level of prostate extracapsular invasion and obtain the level of prostate extracapsular invasion, and inputting the location of the prostate lesion and the level of prostate extracapsular invasion into the third prediction model to predict whether it will be upgraded or not.

[0013] Optionally, the target detection network is trained using image data of different prostate lesion locations;

[0014] Optionally, the target detection network may employ one or more of the following: YOLO V5 / V6 / V7, FCN, FastRCNN, DETR, GhostNetV2;

[0015] Optionally, the recognition network is trained using different levels of extracapsular invasion of the prostate and normal prostate imaging data;

[0016] Optionally, the recognition network may employ one or more of the following: convolutional neural network, generative adversarial neural network, FPN, Mask RCNN, or 3D ResNet-50.

[0017] The recognition network includes an input layer, a basic convolutional module, a multi-scale convolutional module, an edge enhancement module, a fusion module, and a prediction output layer. Prostate images are sequentially processed through the input layer, the basic convolutional layer, the multi-scale convolutional module, and the edge enhancement module to obtain edge enhancement features. The centerline length (CCL) of the contact area in the prostate image is calculated to obtain the CCL. The CCL is mapped to spatial attention weights, and these spatial attention weights are input to the fusion module along with the edge enhancement features to obtain fused features. The fused features are then used by the prediction output layer to obtain the prostate capsule invasion level.

[0018] Optionally, the multi-scale convolutional module includes L sub-modules, where L is a natural number greater than or equal to 1. The sub-modules are connected in series. The output data of the basic convolutional layer passes through the L sub-modules in sequence and is then output to the edge enhancement module. Each sub-module includes a feature extraction layer and an ECA attention module. The output data of the basic convolutional layer passes through the feature extraction layer and the ECA attention module in sequence to obtain attention features. The attention features and the input data are fused through residual connections and then output to the next module.

[0019] Optionally, the feature extraction layer includes a parallel local feature extraction module and a multi-scale feature extraction module. The output data of the basic convolutional layer is processed by the local feature extraction module and the multi-scale feature extraction module to obtain local features and multi-scale features, respectively. The local features and multi-scale features are then fused and input into the ECA attention module.

[0020] Optionally, the edge enhancement module is a BAM module, and the output data of the multi-scale convolution module is enhanced by the BAM module to obtain edge enhancement features.

[0021] The calculation process of CCL is as follows: acquire prostate image data; detect and segment the capsule contact area of ​​the prostate image to obtain the capsule contact area; and calculate the CCL scalar feature by curve integral of the capsule contact area.

[0022] Optionally, the spatial attention weights are obtained by mapping the CCL scalar features to spatial channel attention weights using an MLP;

[0023] Optionally, the fusion module performs gated fusion of the spatial attention weights and the edge enhancement features to obtain fused features.

[0024] The method also includes clinical data, acquiring clinical data of the subject, including one or more of the following: age, tPSA, fPSA, F / T, PSAD, and number of PBCs; inputting the clinical data and the location of the prostate lesion into a fourth prediction model to obtain a prediction result of whether it will be upgraded or not.

[0025] Optionally, clinical data along with the degree of extracapsular invasion of the prostate, PCa diameter, and PI-RADS v2.1 score can be input into a fifth prediction model to obtain a prediction result of whether the prostate will be upgraded or not.

[0026] The purpose of this invention is to provide a computer program product that includes a computer program or instructions, which are executed by a processor to implement the above-mentioned method for predicting postoperative score upgrades of prostate cancer with Gleason scores of 3+3 and 3+4 after radical prostatectomy.

[0027] The purpose of this invention is to provide a computer device comprising a memory, a processor, and a computer program or instructions stored in the memory, wherein the computer program or instructions are executed by the processor to implement the above-mentioned method for predicting postoperative score upgrades of prostate cancer with Gleason scores of 3+3 and 3+4 after puncture.

[0028] The purpose of this invention is to provide a computer-readable storage medium having a computer program or instructions stored thereon, which are executed by a processor to implement the above-mentioned method for predicting postoperative score upgrades of prostate cancer with puncture Gleason scores of 3+3 and 3+4.

[0029] Advantages of this invention:

[0030] 1. To address the issue of inconsistency between pGS and bGS affecting prostate cancer treatment, a predictive method for post-radical prostatectomy score escalation based on mpMRI bGS scores of 3+3 and 3+4 is proposed. This method predicts Gleason score escalation based on the location of the prostate cancer lesion and extracapsular invasion of the prostate. This method has good clinical value and helps doctors predict the Gleason score level after radical prostatectomy, thereby providing appropriate treatment plans.

[0031] 2. The degree of extracapsular invasion of the prostate is related to the prediction of Gleason score escalation after radical prostatectomy. To improve the accuracy of prediction, a recognition network is proposed for the identification and prediction of the degree of extracapsular invasion of the prostate. The recognition network includes an input layer, a basic convolutional module, a multi-scale convolutional module, an edge enhancement module, a fusion module, and a prediction output layer. Among them, the multi-scale convolutional module improves the detection ability of infiltrated areas and small lesions through multi-scale dilated convolution and channel attention. The edge enhancement module integrates prior knowledge of edge gradients to enhance the morphological features of the capsule, which helps the recognition network to extract and recognize smile features and improve the accuracy of identifying the degree of extracapsular invasion of the prostate.

[0032] 3. In response to the different levels of extracapsular invasion of the prostate, this invention incorporates the feature of the centerline length (CCL) of the contact area into the identification network. By combining the geometric measurement information of the CCL with the texture and morphological features of the image, the model's ability to perceive the anatomical standards of grading is enhanced. By jointly learning the CCL and capsule morphological features, the problem of misjudgment caused by a single feature is solved. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 This is a schematic diagram of the method for predicting post-radical prostatectomy score upgrades for Gleason biopsy scores of 3+3 and 3+4, provided in an embodiment of the present invention.

[0035] Figure 2 A schematic diagram of the post-radical prostatectomy score upgrade prediction system for Gleason biopsy scores of 3+3 and 3+4 provided in an embodiment of the present invention;

[0036] Figure 3 A schematic diagram of a prostate cancer post-radical prostatectomy score upgrade prediction device with Gleason scores of 3+3 and 3+4 provided in an embodiment of the present invention;

[0037] Figure 4 The Gleason score provided for this embodiment of the invention was upgraded from 3+3 at the time of biopsy to 4+3 after total resection;

[0038] Figure 5 The Gleason score provided for this embodiment of the invention was upgraded from 3+4 at the time of biopsy to 4+5 after total resection;

[0039] Figure 6 The AUC results of GS upgrade after total resection provided by the clinical model, mpMRI model and combined diagnostic model in the embodiments of the present invention;

[0040] Figure 7 This is a network structure diagram of the identification network provided in an embodiment of the present invention;

[0041] Figure 8 This is a network structure diagram of the multi-scale convolution module provided in an embodiment of the present invention. Detailed Implementation

[0042] To enable those skilled in the art to better understand the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0043] In some of the processes described in the specification, claims, and accompanying drawings of this invention, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as S101, S102, etc., are merely used to distinguish different operations and do not represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types.

[0044] Figure 1 The schematic diagram of the method for predicting post-radical prostatectomy score upgrades for Gleason biopsy scores of 3+3 and 3+4 provided in this embodiment of the invention specifically includes:

[0045] S1: Based on the acquisition of prostate mpMRI imaging data and preoperative biopsy data of the subject;

[0046] In one specific embodiment, the present invention collected clinical and MRI data of 401 patients with prostate disease from November 2014 to December 2022. The inclusion criteria for patients were as follows: (1) complete MRI and laboratory test data; (2) confirmed diagnosis by ultrasound-guided prostate biopsy and RP; (3) bGS of 3+3 or 3+4; (4) no treatment received before prostate biopsy and RP. Exclusion criteria: (1) patients whose bGS was not 3+3 or 3+4 (n = 230); (2) patients who received treatment before prostate biopsy and RP (n = 14); (3) patients with incomplete clinical or imaging data (n = 12).

[0047] Clinical pathological data were derived from patient medical records, including prostate volume, total prostate-specific antigen (tPSA), free PSA (fPSA), free / total PSA ratio (F / T), PSA density (PSAD), number of positive biopsies (PBCs), time interval between puncture and RP, bGS, and pGS. PSA values ​​were measured within 8 weeks prior to the mpMRI examination. An ultrasound-guided transperineal prostate biopsy was performed by a urologist. The urologist first performed 11 systematic punctures, followed by at least one additional cognitive-targeted puncture. The cognitive-targeted puncture involved the urologist reviewing the MRI results and using this data to target the most prominent nodules under ultrasound guidance. The entire resection specimen was sectioned from apex to base at 3-5 mm intervals, and the corresponding PCa boundaries were delineated on each section. A pathologist with 10 years of experience simultaneously reported the pathological scores of both the biopsy and RP specimens. PCa was graded histopathologically using the Gleason grading system.

[0048] In one specific embodiment, MRI scans were performed using a GE 750 3.0-T scanner (GE Healthcare, Milwaukee, WI, USA). The scan sequences included axial T1-weighted images (T1WI) and T2-weighted images (T2WI), coronal and sagittal T2WI, diffusion-weighted imaging (DWI), and dynamic contrast-enhanced imaging (DCE). The contrast agent was gadolinium, at a flow rate of 3 mL / s and a concentration of 0.1 mmol / kg.

[0049] Two radiologists (reader 1: 5 years of experience; reader 2: 14 years of experience) independently analyzed the images under double-blind conditions. When the two radiologists disagreed, they reached a consensus. A third radiologist (reader 3: 4 years of experience) matched the lesions on the MRI images with the pathology report. The diameter of the prostate capsule (PCa), PCa location (peripheral / transitional zone, base / middle / apex), extracapsular invasion (EPE) grade, and PI-RADS v2.1 were measured on mpMRI.

[0050] On axial T2WI, the peripheral zone and transition zone can be distinguished, and both can be involved when the lesion is large. On sagittal T2WI, the prostate is evenly divided into three parts: base, middle and apex. The EPE grading criteria are as follows: (1) Grade 0: no obvious capsular contact, or curve contact length (CCL) < 15 mm; (2) Grade 1: CCL ≥ 15 mm, or irregular or protruding capsule; (3) Grade 2: CCL ≥ 15 mm, and irregular or protruding capsule; (4) Grade 3: obvious capsule breakthrough. The lesion images of each sequence were scored by radiologists using mpMRI and the PI-RADS v2.1 score was calculated. When the PI-RADS score ≥ 4 points, the lesion was considered to be PCa. When the PI-RADS score < 3 points, the location of PCa could not be obtained and was expressed as N / A. When multiple lesions were present, the lesion with the highest pGS was used for statistical analysis. When pGS are equal, the largest lesion is selected for statistical analysis.

[0051] In one specific embodiment, this invention included 145 patients with prostate disease. 83 patients had a bGS of 3+3, and 62 patients had a bGS of 3+4. Of the 145 patients, 81 (55.9%) experienced GS upgrade, and 64 (44.1%) did not. 53 (63.9%) patients with a bGS of 3+3 had a Gleason score upgrade to 3+4 or higher after total resection, and 28 (45.2%) patients with a bGS of 3+4 had a Gleason score upgrade to 4+3 or higher after total resection. The patients' ages ranged from 46 to 80 years, with a median tPSA value of 7.67 ng / mL. The main clinical manifestations were elevated PSA, dysuria, urinary frequency, and urgency. A total of 51 (35.2%) patients developed end-stage renal disease (EPE). PI-RADS v2.1 grading: 118 lesions scored 4 or 5, 15 lesions scored 3, and 12 lesions scored 2. The mpMRI readings provided by the two radiologists showed a high degree of agreement (κ = 0.752).

[0052] Table 1 shows the differences between the upgraded GS group and the non-upgraded GS group. The proportion of lesions located at the base was higher in the upgraded GS group than in the non-upgraded GS group, and the grade of EPE was higher in the upgraded GS group than in the non-upgraded GS group (P < 0.05 for both). In the upgraded GS group, 24 cases (29.6%) of prostate lesions were located at the base, while only 6 cases (9.4%) of lesions were located at the base in the non-upgraded GS group. Of the 81 upgraded GS patients, 19 cases (23.5%) had grade 1 EPE, 15 cases (18.5%) had grade 2 EPE, 2 cases (2.4%) had grade 3 EPE, and 45 cases (55.6%) had no EPE. Of the 64 non-upgraded GS patients, 11 cases (17.2%) had grade 1 EPE, 4 cases (6.2%) had grade 2 EPE, and 49 cases (76.6%) had no EPE.

[0053] Figure 4 This study presents a 62-year-old male patient whose Gleason score improved from 3+3 at biopsy to 4+3 post-RP. Axial T2WI images showed focal hypointense areas at the base of the transition zone, with an EPE grade of 2 and a PI-RADS v2.1 score of 5. Figure 5 This study presents a 66-year-old male patient whose Gleason score improved from 3+4 at biopsy to 4+5 post-RP. Axial T2WI images showed focal hypointense areas at the basal peripheral zone, with an EPE grade of 2 and a PI-RADS v2.1 score of 5.

[0054] Table 1

[0055]

[0056] Table 2

[0057]

[0058] Univariate and multivariate logistic regression analyses were used to analyze the influencing factors of GS progression after RP, and the results are shown in Table 2. To evaluate the predictive performance of clinical and mpMRI variables on GS progression after RP, we constructed ROC curves for clinical factors (age, number of PBCs, time from puncture to RP, tPSA, fPSA, F / T, prostate volume, PSAD), ROC curves for mpMRI results (PCa diameter, PCa location, EPE grade, and PI-RADS v2.1), and a combined diagnostic model including all clinical and mpMRI variables (…). Figure 6 (As shown). The AUC values ​​for clinical factors and mpMRI findings were 0.674 and 0.697, respectively. The AUC of the combined diagnostic model was 0.777, which was greater than that of clinical factors (P = 0.009).

[0059] In one specific embodiment, the present invention demonstrates that patients with tumors located at the base are more likely to experience GS upgrade after RP. The GS upgrade group is more likely to experience EPE, and the EPE grade is higher than that in the non-GS upgrade group. This confirms that the GS upgrade group has a worse postoperative prognosis, and therefore these patients are not suitable for active surveillance, as opposed to the non-GS upgrade group.

[0060] AS has become the standard of care for low-risk PCa, avoiding overtreatment of PCa; AS is recommended for patients with GS ≤ 6. In some cases, patients with GS 3 + 4 are candidates for AS, while in others, aggressive treatment is necessary. PCa with GS 3 + 4 has a worse prognosis than PCa with GS 3 + 3, but a better prognosis than PCa with GS 4 + 3.

[0061] S2: The Gleason score for prostate cancer in the subject is obtained based on preoperative biopsy data;

[0062] In one embodiment, the prostate score is 3+3 or 3+4.

[0063] S3: The location of the prostate lesion is obtained based on the image data; when the prostate lesion is located at the bottom, the postoperative escalation is determined based on the Gleason score of prostate cancer to obtain the prediction result.

[0064] In one embodiment, S3 is replaced by: obtaining the location of the prostate lesion based on the image data, and inputting the location of the prostate lesion into the first prediction model to obtain a prediction result of whether it will be upgraded or not.

[0065] Optionally, the first prediction model may employ one or more of the following: SVM, XGBoost / LightGBM, convolutional neural network, or Transformer.

[0066] In one embodiment, the method further includes the prostate capsule invasion level, which is obtained based on the imaging data. When the prostate capsule invasion level is greater than or equal to level 2 and the prostate lesion is located at the bottom, the postoperative escalation is determined based on the prostate scoring level to obtain a prediction result.

[0067] In one embodiment, the method further includes the prostate capsule invasion grade, PCa diameter, and PI-RADS v2.1 score; the PCa diameter and PI-RADS v2.1 score are obtained based on the image data; and the prostate lesion location, prostate capsule invasion grade, PCa diameter, and PI-RADS v2.1 score are input into a second prediction model to predict whether the lesion will be upgraded or not.

[0068] In one embodiment, the method further includes clinical data, acquiring clinical data of the subject, including one or more of the following: age, tPSA, fPSA, F / T, PSAD, and number of PBCs; inputting the clinical data and the location of the prostate lesion into a fourth prediction model for prediction to obtain a prediction result of whether the lesion will be upgraded or not.

[0069] In one embodiment, clinical data, along with the degree of extracapsular invasion of the prostate, PCa diameter, and PI-RADS v2.1 score, are input into a fifth prediction model to obtain a prediction result of whether the prostate will be upgraded or not.

[0070] In one embodiment, S3 is replaced by: inputting the image data into a target detection network to detect and obtain the location of the prostate lesion, and simultaneously inputting the image data into a recognition network to determine the level of prostate extracapsular invasion and obtain the level of prostate extracapsular invasion, and inputting the location of the prostate lesion and the level of prostate extracapsular invasion into a third prediction model to predict whether it will be upgraded or not.

[0071] In one embodiment, the target detection network is trained using image data of different prostate lesion locations;

[0072] Optionally, the target detection network may employ one or more of the following: YOLO V5 / V6 / V7, FCN, FastRCNN, DETR, GhostNetV2.

[0073] In one embodiment, the identification network is trained using different levels of extracapsular prostate invasion and normal prostate imaging data;

[0074] Optionally, the recognition network may employ one or more of the following: convolutional neural network, generative adversarial neural network, FPN, Mask RCNN, or 3D ResNet-50.

[0075] In another embodiment, the recognition network includes an input layer, a basic convolutional module, a multi-scale convolutional module, an edge enhancement module, a fusion module, and a prediction output layer. The prostate image is sequentially passed through the input layer, the basic convolutional layer, the multi-scale convolutional module, and the edge enhancement module to obtain edge enhancement features. The centerline length (CCL) of the contact area in the prostate image is calculated to obtain the CCL. The CCL is mapped to spatial attention weights. The spatial attention weights and the edge enhancement features are input to the fusion module for fusion to obtain fused features. The fused features are used to obtain the prostate capsule invasion level through the prediction output layer.

[0076] Optionally, the multi-scale convolutional module includes L sub-modules, where L is a natural number greater than or equal to 1. The sub-modules are connected in series. The output data of the basic convolutional layer passes through the L sub-modules in sequence and is then output to the edge enhancement module. Each sub-module includes a feature extraction layer and an ECA attention module. The output data of the basic convolutional layer passes through the feature extraction layer and the ECA attention module in sequence to obtain attention features. The attention features and the input data are fused through residual connections and then output to the next module.

[0077] In one embodiment, the feature extraction layer includes a parallel local feature extraction module and a multi-scale feature extraction module. The output data of the basic convolutional layer is processed by the local feature extraction module and the multi-scale feature extraction module to obtain local features and multi-scale features, respectively. The local features and multi-scale features are then fused and input into the ECA attention module.

[0078] In one embodiment, the edge enhancement module is a BAM module, and the output data of the multi-scale convolution module is enhanced by the BAM module to obtain edge enhancement features.

[0079] In one embodiment, the calculation process of CCL is as follows: acquiring prostate image data; detecting and segmenting the capsule contact area of ​​the prostate image to obtain the capsule contact area; and calculating the CCL scalar feature by curve integral of the capsule contact area.

[0080] In one embodiment, the spatial attention weights are obtained by mapping the CCL scalar features to spatial channel attention weights using an MLP.

[0081] In one embodiment, the fusion module performs gated fusion of spatial attention weights and edge enhancement features to obtain fused features.

[0082] In one specific embodiment, the main architecture of the identified network is as follows: Figure 7 As shown, it includes an input layer, a basic convolutional module, two multi-scale convolutional modules, an edge enhancement module, a CCL calculation module, a CCL mapping module, a fusion module, and an output layer. These modules form parallel branches to obtain edge enhancement features and CCL spatial attention weights, which are then fused by the fusion module to predict the prostate capsule invasion level.

[0083] The basic convolutional module includes a 3D convolutional layer and a max pooling layer. Prostate image data (input dimension 1×128×128×128) is input through the input layer and then processed by the convolutional and max pooling layers in the basic convolutional layer to obtain the output features of this module (64×64×64×64). The output data of this module is then input into the multi-scale convolutional module.

[0084] The network structure of the multi-scale convolutional module 1 is as follows: Figure 8 As shown:

[0085] Input feature map (64×64×64×64);

[0086] Branch 1: Standard 3D convolution, sequentially passing through 3×3×3 convolution (dilation rate = 1) - BN layer - ReLU;

[0087] Branch 2: Multi-scale dilated convolution, including 3×3×3 dilated convolution (dilation rate=3)-BN layer-ReLU and 3×3×3 dilated convolution (dilation rate=5)-BN layer-ReLU; the features of the multi-scale dilated convolution are concatenated (along the channel axis) and dimensionality reduced by 1×1×1 convolution to output 128 channels;

[0088] The features from branch 1 and branch 2 are fused and fed into the ECA attention module. Global average pooling is performed first, followed by channel attention calculation to complete channel weighting and obtain attention features. The input feature map and attention features are then added and fused through residual connection and fed into the ReLU layer for activation and 3D max pooling (kernel 2×2×2) to obtain the output features of the module.

[0089] The network structure of multi-scale convolution module 2 is the same as that of multi-scale convolution module 1. Multi-scale convolution module 1 uses a moderate expansion rate (1 / 3 / 5) to match the anatomical structure of 5-10mm in the central region of the prostate, focusing on distinguishing the tumor infiltration patterns of the peripheral zone and the transition zone. Multi-scale convolution module 2 increases the expansion rate (2 / 4 / 6) to cover the extracapsular regions such as the seminal vesicle / Diesselbach's fascia, and strengthens the characteristics of the capsule contact surface.

[0090] After the multi-scale convolution module outputs features, it is fed into the BAM module. The BAM module generates a capsule edge gradient map through the edge gradient extraction layer, and then weights and fuses the gradient map with the input features of the BAM module to obtain edge enhancement features, which are used to enhance the capsule contour features.

[0091] In one specific embodiment, CCL calculation is performed as follows: Based on the prostate MRI sequence, the capsule contact area (features) is segmented using Sobel edge detection and morphological closure operation. Based on the capsule contact area, the length of the centerline of the contact area is calculated using a curve integral algorithm to generate millimeter-level scalar features. The CCL scalar features are mapped to spatial channel attention weights using MLP and then gated and fused with edge enhancement features to enhance the CCL-related areas.

[0092] The CCL length is mapped to a spatial weight Fout, which is then weighted and fused with the features of the capsule contact region, using the following formula:

[0093] Fout = Fin·σ(MLP(CCL))

[0094] Where σ is the Sigmoid function, MLP converts the CCL scalar into a spatial attention map, and Fin is the feature of the capsule contact region.

[0095] Edge enhancement features are fused with CCL spatial weights to obtain fused features, and finally the final level prediction result is obtained through the prediction output layer (composed of global average pooling layer and FC layer).

[0096] The loss function is optimized as follows:

[0097] L=0.7·LCE+0.3·LMSE+0.2·Ledge

[0098] LCE represents cross-entropy loss, LMSE represents constraint on the consistency of CCL prediction with the gold standard (error ≤ 0.3cm), and Ledge represents edge consistency loss. The Sobel gradient map is used to constrain the topological relationship between the predicted lesion boundary and the CCL region. The weights are dynamically adjusted. When the CCL prediction error is > 0.5cm, the weight of LMSE is automatically increased to strengthen the model's learning of the critical value.

[0099] The present invention also discloses a computer program product or system, including a computer program that, when executed by a processor, implements the above-described method steps for predicting post-radical prostatectomy score upgrades for Gleason biopsy scores of 3+3 and 3+4.

[0100] Figure 2 The schematic diagram of the post-radical prostatectomy score upgrade prediction system for Gleason biopsy scores of 3+3 and 3+4 provided in this embodiment of the invention specifically includes:

[0101] Acquisition module: Based on acquiring the prostate mpMRI image data and preoperative biopsy data of the subject;

[0102] The rating module: derives the Gleason score for prostate cancer in the subject based on preoperative biopsy data;

[0103] Prediction module: Based on the image data, the location of the prostate lesion is obtained; when the prostate lesion is located at the bottom, the postoperative escalation is determined based on the Gleason score of prostate cancer, and a prediction result is obtained.

[0104] Figure 3 The schematic diagram of the prostate cancer post-radical prostatectomy score upgrade prediction device with Gleason biopsy scores of 3+3 and 3+4 provided in this embodiment of the invention specifically includes:

[0105] A memory and a processor; the memory is used to store program instructions; the processor is used to invoke the program instructions, when any of the program instructions are executed, the above-mentioned method for predicting the score upgrade after radical prostatectomy for Gleason scores of 3+3 and 3+4.

[0106] The present invention also discloses a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, any one of the above-described methods for predicting post-radical prostatectomy score upgrades for Gleason scores of 3+3 and 3+4 obtained through biopsy is provided.

[0107] The verification results of this verification embodiment show that assigning inherent weights to indications can improve the performance of this method compared to the default settings. Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. In the several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for example, the division of units is merely a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling or direct coupling or communication connection may be through some interfaces, indirect coupling or communication connection of devices or units, and may be electrical, mechanical, or other forms. The units described as separate components may or may not be physically separated; the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of this embodiment. Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units. Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. This program can be stored in a computer-readable storage medium, which may include: read-only memory (ROM), random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0108] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0109] The computer device provided by the present invention has been described in detail above. For those skilled in the art, there will be changes in the specific implementation and application scope based on the ideas of the embodiments of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for predicting post-radical prostatectomy upgrading of a prostate cancer with a Gleason score of 3+3 and 3+4, characterized by, The method comprises the following steps: S1, based on the obtained prostate mpMRI image data of the subject, preoperative biopsy data; S2, based on the preoperative biopsy data, obtaining the prostate cancer Gleason score grade of the subject; S3, inputting the image data into a target detection network for detection to obtain the position of the prostate lesion, and inputting the image data into an identification network for identification to obtain the extraprostatic extension grade of the prostate capsule, inputting the position of the prostate lesion and the extraprostatic extension grade of the prostate capsule into a third prediction model for prediction to obtain the prediction result of upgrading or not upgrading; when the extraprostatic extension grade of the prostate capsule is greater than or equal to two levels and the position of the prostate lesion is at the bottom, the postoperative upgrading is determined based on the prostate Gleason score grade to obtain the prediction result; The target detection network is obtained by training different prostate lesion position image data; The target detection network adopts one or more of the following: YOLO V5 / V6 / V7, FCN, Fast RCNN, DETR, and GhostNetV2; The identification network is obtained by training different levels of extraprostatic extension grade of the prostate capsule and normal prostate image data; The identification network comprises an input layer, a basic convolution module, a multi-scale convolution module, an edge enhancement module, a fusion module, and a prediction output layer; the prostate image sequentially passes through the input layer, the basic convolution layer, the multi-scale convolution module, and the edge enhancement module to obtain the edge enhancement feature, the center line length CCL of the contact area in the prostate image is calculated to obtain CCL, the CCL is mapped to a spatial attention weight, the spatial attention weight and the edge enhancement feature are input into the fusion module for fusion to obtain a fusion feature, and the fusion feature is input into the prediction output layer to obtain the extraprostatic extension grade of the prostate capsule.

2. The method of predicting post-radical prostatectomy grade upgrading for prostate cancer with a Gleason score of 3+3 and 3+4 according to claim 1, characterized in that, The prostate cancer Gleason score grade is 3+3 or 3+4.

3. The method of predicting post-radical prostatectomy grade upgrading of prostate cancer with a Gleason score of 3+3 and 3+4 according to claim 1, characterized by, The method further comprises the extraprostatic extension grade of the prostate capsule, the PCa diameter, and the PI-RADS v2.1 score, the PCa diameter and the PI-RADS v2.1 score are obtained based on the image data, and the position of the prostate lesion, the extraprostatic extension grade of the prostate capsule, the PCa diameter, and the PI-RADS v2.1 score are input into a second prediction model for prediction to obtain the prediction result of upgrading or not upgrading.

4. The method of predicting post-radical prostatectomy grade upgrading of prostate cancer with a Gleason score of 3+3 and 3+4 according to claim 1, characterized by, The multi-scale convolution module comprises L sub-modules, L is a natural number greater than or equal to 1, the sub-modules are connected in series, and the output data of the basic convolution layer is output to the edge enhancement module after sequentially passing through the L sub-modules; the sub-module comprises a feature extraction layer and an ECA attention module, the output data of the basic convolution layer passes through the feature extraction layer and the ECA attention module of the sub-module in sequence to obtain an attention feature, and the attention feature and the input data are fused through residual connection and then output to the next module.

5. The method of predicting post-radical prostatectomy grade upgrading of prostate cancer with a Gleason score of 3+3 and 3+4 according to claim 4, characterized by, The feature extraction layer comprises parallel local feature extraction modules and multi-scale feature extraction modules, and the output data of the basic convolution layer are respectively input into the local feature extraction modules and the multi-scale feature extraction modules to obtain local features and multi-scale features.

6. The method of predicting post-radical prostatectomy grade upgrading of prostate cancer with a Gleason score of 3+3 and 3+4 according to claim 1, characterized by, The edge enhancement module is a BAM module, and the output data of the multi-scale convolution module are input into the BAM module to obtain edge enhancement features through edge gradient enhancement.

7. The method of predicting post-radical prostatectomy grade upgrading of prostate cancer with a Gleason score of 3+3 and 3+4 according to claim 1, characterized by, The CCL calculation process comprises: acquiring prostate image data; performing capsule contact area detection and segmentation on the prostate image to obtain a capsule contact area; and performing curve integral calculation on the capsule contact area to obtain a CCL scalar feature.

8. The method of predicting post-radical prostatectomy grade upgrading of prostate cancer with a Gleason score of 3+3 and 3+4 according to claim 1, characterized by, The spatial attention weight is obtained by mapping the CCL scalar feature to a spatial channel attention weight through an MLP.

9. The method of predicting post-radical prostatectomy grade upgrading of prostate cancer with a Gleason score of 3+3 and 3+4 according to claim 1, characterized by, The fusion module performs gated fusion on the spatial attention weight and the edge enhancement features to obtain fusion features.

10. The method of predicting post-radical prostatectomy grade upgrading of prostate cancer with a Gleason score of 3+3 and 3+4 according to claim 1, characterized in that, The method further comprises clinical data, acquiring clinical data of a to-be-tested person, including one or more of the following: age, tPSA, fPSA, F / T, PSAD, and PBC quantity; inputting the clinical data and the position of the prostate lesion into a fourth prediction model to obtain a prediction result of upgrading or not upgrading.

11. The method of predicting post-radical prostatectomy grade upgrading of prostate cancer with a Gleason score of 3+3 and 3+4 according to claim 10, characterized in that, Clinical data, prostate capsule extracapsular invasion grade, PCa diameter, and PI-RADS v2.1 score are input into a fifth prediction model to obtain a prediction result of upgrading or not upgrading.

12. A computer program product comprising a computer program or instructions embodied therein, characterized in that, The computer program or instructions are executed by the processor to implement the mpMRI post-prostate cancer radical surgery upgrading prediction method of any one of claims 1-11.

13. A computer device comprising a memory, a processor, and a computer program or instructions stored on the memory, wherein, The computer program or instructions are executed by the processor to implement the mpMRI post-prostate cancer radical surgery upgrading prediction method of any one of claims 1-11.

14. A computer readable storage medium having stored thereon a computer program or instructions, characterized in that, The computer program or instructions are executed by the processor to implement the mpMRI post-prostate cancer radical surgery upgrading prediction method of any one of claims 1-11.

Citation Information

Patent Citations

  • Method, device and program product for predicting Gleason score upgrading and degrading after radical prostatectomy

    CN119132512A