Detection method and device for screening CDKN2A homozygous deletion in central nervous system glioma, equipment and storage medium
By combining digital panoramic scanning and deep learning models with ROC curve analysis, a rapid and accurate screening for homozygous CDKN2A deletions in central nervous system gliomas was achieved, solving the problems of complex and costly detection in existing technologies and improving screening efficiency and accuracy.
Patent Information
- Application Number
- CN202511040884.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-28
- Publication Date
- 2025-12-12
AI Technical Summary
Existing technologies are complex, time-consuming, and costly in detecting homozygous CDKN2A deletions in central nervous system gliomas, making them unsuitable for large-scale initial screening. Furthermore, existing MTAP immunohistochemical staining analysis suffers from inconsistent scoring standards and strong subjectivity.
Digital panoramic scanners are used to acquire sliced digital images. Combined with deep learning models, MTAP signals are automatically identified. The optimal cutoff value is determined through ROC curve statistical analysis, enabling rapid and accurate screening of CDKN2A homozygous deletions.
It improves the screening efficiency of CDKN2A homozygous deletion, reduces the blind spots and costs of molecular testing, facilitates clinical application, and combines the professional experience of pathologists with the high-throughput and objective analysis advantages of machine learning.
Smart Images
Figure CN121120490A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the fields of tumor molecular diagnostics and pathology, and in particular to a method, apparatus, equipment and storage medium for screening for homozygous deletion of CDKN2A in central nervous system gliomas. Background Technology
[0002] The CDKN2A gene encodes the p16 protein, an important tumor suppressor in cell cycle regulation. Its homozygous deletion is a common genetic abnormality in various malignant tumors (including gliomas), typically indicating poor prognosis and increased invasiveness. Accurate detection of CDKN2A homozygous deletion is crucial for the classification, prognosis, and individualized treatment planning of glioma patients.
[0003] Currently, clinical detection of CDKN2A deletion mainly relies on molecular technologies such as fluorescence in situ hybridization (FISH), real-time quantitative PCR, or next-generation sequencing (NGS). Although these technologies are highly sensitive, they are complex to operate, time-consuming, and costly, making them unsuitable for large-scale initial screening.
[0004] The MTAP gene is located in the adjacent region of CDKN2A. Given that its expression loss is highly correlated with homozygous loss of CDKN2A, immunohistochemical staining of MTAP protein has become a potential alternative screening method. Previous studies have shown that MTAP protein expression loss can serve as an indirect marker of homozygous loss of CDKN2A. However, existing literature reports problems such as inconsistent scoring criteria, high subjectivity, and a lack of accurate cutoff values, preventing this technology from meeting the requirements for widespread clinical application.
[0005] Therefore, developing an objective, standardized, and efficient MTAP immunohistochemical staining analysis method, combined with advanced machine learning image processing technology, and scientifically determining the optimal score cutoff value for identifying homozygous CDKN2A deletion, has significant clinical application value. Summary of the Invention
[0006] This application provides a method, apparatus, equipment, and storage medium for screening homozygous deletions of CDKN2A in central nervous system gliomas. The optimal cutoff value is determined through ROC curve statistical analysis, enabling rapid and accurate screening of homozygous deletions of CDKN2A in central nervous system gliomas.
[0007] In a first aspect, this application provides a method for detecting homozygous deletion of CDKN2A in central nervous system gliomas, comprising:
[0008] Obtain sliced digital images;
[0009] Calculate the first score of the sliced digital image;
[0010] The sliced digital image is input into a pre-trained deep learning model to automatically identify and score the MATP signal in the image, thus obtaining a second score.
[0011] The first score and the second score were compared with the gold standard of homozygous deletion of CDKN2A gene in the corresponding samples, and the first ROC curve and the second ROC curve were plotted.
[0012] Based on the first ROC curve and the second ROC curve, a first optimal cutoff threshold and a second optimal cutoff threshold are determined respectively. The first optimal cutoff threshold and the second optimal cutoff threshold are used as the criteria for screening for homozygous deletion of CDKN2A, thereby achieving the screening of homozygous deletion of CDKN2A.
[0013] In one possible design, sliced digital images are obtained as follows:
[0014] The stained images were scanned using a digital panoramic scanner to obtain digital images of the slides; wherein, the stained images were obtained by immunohistochemical staining of MTAP protein on paraffin sections of multiple tissues from central nervous system gliomas.
[0015] In one possible design, the first score of the sliced digital image is calculated using the following formula:
[0016]
[0017] In the formula, H represents the first score, i represents the MTAP staining intensity value, and I... i The value of i represents the percentage of tumor area with an MTAP staining intensity of i, and N represents the maximum MTAP staining intensity.
[0018] In one possible design, the sliced digital image is input into a pre-trained deep learning model to automatically identify and score the MATP signals in the image, and the second score is calculated using the following formula:
[0019] AOD = IntDen / Count
[0020] In the formula, AOD represents the second score, IntDen represents the integrated optical density, and Count is the pixel count.
[0021] In one possible design, the deep learning model is an image recognition model based on a deep convolutional neural network.
[0022] In one possible design, the first optimal cutoff threshold and the second optimal cutoff threshold are used as the criteria for screening for homozygous deletions of CDKN2A, thereby achieving the screening of homozygous deletions of CDKN2A, including:
[0023] If the first score of the sliced digital image is less than or equal to the first optimal truncation threshold or the second score is less than or equal to the second optimal truncation threshold, the sample corresponding to the sliced digital image is determined to be a CDKN2A homozygous missing value.
[0024] In one possible design, the methods for determining the first optimal cutoff threshold and the second optimal cutoff threshold based on the first ROC curve and the second ROC curve respectively include the Youden index method, the geometric distance method, the cost-benefit ratio minimization method, and the Youden index method.
[0025] Secondly, this application provides a detection device for screening CDKN2A homozygous deletion in central nervous system gliomas, comprising:
[0026] The image acquisition module is configured to acquire sliced digital images;
[0027] The first scoring calculation module is configured to calculate a first score for the sliced digital image;
[0028] The second scoring calculation module is configured to input the sliced digital image into a pre-trained deep learning model, automatically identify and score the MATP signal in the image, and obtain a second score.
[0029] The ROC curve generation module is configured to compare the first score and the second score with the corresponding sample's CDKN2A gene homozygous deletion gold standard, and plot the first ROC curve and the second ROC curve.
[0030] The optimal cutoff threshold determination module is configured to determine a first optimal cutoff threshold and a second optimal cutoff threshold based on the first ROC curve and the second ROC curve, respectively, and use the first optimal cutoff threshold and the second optimal cutoff threshold as the criteria for screening for homozygous deletion of CDKN2A, thereby realizing the screening for homozygous deletion of CDKN2A.
[0031] Thirdly, embodiments of this application provide an electronic device, including: at least one processor and a memory; the memory stores computer execution instructions; the at least one processor executes the computer execution instructions stored in the memory, causing the at least one processor to perform the detection method for screening CDKN2A homozygous deletion in central nervous system gliomas as described in the first aspect and various possible designs of the first aspect.
[0032] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions. When a processor executes the computer-executable instructions, it implements the detection method for screening CDKN2A homozygous deletions in central nervous system gliomas as described in the first aspect and various possible designs of the first aspect.
[0033] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the detection method for screening homozygous deletions of CDKN2A in central nervous system gliomas as described in the first aspect and various possible designs of the first aspect.
[0034] The method, apparatus, equipment, and storage medium provided in this application for screening CDKN2A homozygous deletions in central nervous system gliomas have at least the following beneficial effects:
[0035] This application combines the professional experience of pathologists with the high-throughput and objective analysis advantages of machine learning; it uses ROC curves to scientifically determine the screening cutoff value, ensuring a balance between screening sensitivity and specificity; it is applicable to routine immunohistochemical slides, requiring no additional complex equipment or high costs, making it easy to promote in clinical practice; it improves the efficiency of CDKN2A homozygous deletion screening and reduces the blind spots and cost burden of molecular testing. Attached Figure Description
[0036] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.
[0037] Figure 1 A flowchart illustrating the detection method for screening CDKN2A homozygous deletions in central nervous system gliomas provided in this application embodiment;
[0038] Figure 2 A first scoring diagram provided for an embodiment of this application;
[0039] Figure 3 A schematic diagram of the first ROC curve provided in an embodiment of this application;
[0040] Figure 4 A schematic diagram of the second ROC curve provided in an embodiment of this application;
[0041] Figure 5 Performance verification diagram of the verification model provided in the embodiments of this application;
[0042] Figure 6 This is a schematic diagram of the detection device for screening CDKN2A homozygous deletion in central nervous system gliomas provided in an embodiment of this application.
[0043] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation
[0044] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.
[0045] The collection, storage, use, processing, transmission, provision, and disclosure of information such as financial data, user data, or medical image data involved in the technical solution of this application all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.
[0046] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.
[0047] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.
[0048] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.
[0049] This application provides a method for detecting homozygous deletions of CDKN2A in central nervous system gliomas. Please refer to [link to relevant documentation]. Figure 1 , Figure 1 The flowchart of the detection method for screening CDKN2A homozygous deletion in central nervous system gliomas provided in the embodiments of this application includes the following steps S10-S50.
[0050] S10: Obtain the sliced digital image.
[0051] In some embodiments, digital images of the slides are obtained by scanning the stained images using a digital panoramic scanner; wherein the stained images are obtained by performing MTAP protein immunohistochemical staining on paraffin sections of multiple tissues from central nervous system gliomas.
[0052] In this embodiment, MTAP protein immunohistochemical staining was performed on paraffin sections of 312 tissues from central nervous system gliomas. A professionally trained pathologist manually performed semi-quantitative scoring on the MTAP staining intensity and positive rate of each section. The scoring criteria included staining intensity of 0-3 and positive cell ratio of 0-100%. The obtained data will be used as the basis for the calculation of the first score in step S20, and digital images of the sections will be acquired using a digital panoramic scanner.
[0053] S20: Calculate the first score of the sliced digital image.
[0054] In some embodiments, the first score of the sliced digital image is calculated using the following formula:
[0055]
[0056] In the formula, H represents the first score, i represents the MTAP staining intensity value, and I... i The value of i represents the percentage of tumor area with an MTAP staining intensity of i, and N represents the maximum MTAP staining intensity.
[0057] In this embodiment, the first score uses the H-score to evaluate the immune expression of 5hmC and MTAP. In an exemplary embodiment, to calculate the H score (range 0-300), H score = 0 × (percentage of cell staining area with intensity value of 0) + 1 × (percentage of cell staining area with intensity value of 1) + 2 × (percentage of cell staining area with intensity value of 2) + 3 × (percentage of cell staining area with intensity value of 3). For statistical purposes, the score is used as a continuous variable or divided in binary: 0-300. An exemplary first score diagram is shown below. Figure 2 As shown, (a)-(d) represent 0 points, 1 point, 2 points, and 3 points, respectively. (See Table 1 for scoring details) Table 1 Scoring Details
[0058]
[0059] The advantage of using the first scoring method mentioned above is that:
[0060] (1) Flexible handling of complex organizational forms
[0061] It can identify tumor heterogeneity (such as focal MTAP-deficient areas) and avoid false negatives / positives caused by interference from normal cells or necrotic areas in tissue sections.
[0062] Adapt to special staining patterns (such as weak positive, abnormal cytoplasmic / nuclear localization), and make a comprehensive judgment based on the experience of the pathologist.
[0063] (2) Robustness to technical errors
[0064] Experimental errors such as uneven staining, detachment, or edge effects can be corrected manually (e.g., by excluding areas where staining failed).
[0065] Differentiate from non-specific background staining (such as residual MTAP expression in macrophages or mesenchymal cells).
[0066] (3) Low cost and rapid deployment
[0067] No specialized equipment is required, making it suitable for small and medium-sized laboratories with limited resources.
[0068] It has a high tolerance for sample quality (such as poor fixation or old FFPE).
[0069] S30: Input the sliced digital image into the pre-trained deep learning model to automatically identify and score the MATP signal in the image, and obtain the second score.
[0070] In some embodiments, digital tile images are acquired using a panoramic scanner (such as Leica Aperio). A deep convolutional neural network model is constructed and trained, taking image patches as input and outputting an MTAP positivity score. The model training set contains sample data with known CDKN2A states. The model score is automatically output as a numerical score, meaning that the final score of the digital tile image can be output through the pre-trained deep learning model.
[0071] In some embodiments, the average optical density is used as the second score. AOD (Average Optical Density) = the average of the optical density (OD) of all pixels in the region = IntDen / Count, where IntDen represents the integrated optical density and Count is the pixel count.
[0072] In some embodiments, the deep learning model is an image recognition model based on a deep convolutional neural network.
[0073] S40: Compare the first score and the second score with the corresponding sample's CDKN2A gene homozygous deletion gold standard, and plot the first ROC curve and the second ROC curve.
[0074] The gold standard for detecting homozygous deletions of the CDKN2A gene can be verified by FISH or NGS. FISH uses fluorescently labeled DNA probes to bind directly to the target gene (CDKN2A), allowing for direct detection of gene deletions at the nuclear level. It is particularly suitable for detecting homozygous deletions (loss of both alleles), avoiding false negatives caused by contamination or amplification bias in PCR or sequencing. NGS (next-generation sequencing) uses high-throughput sequencing technology to detect copy number variations (CNVs), ensuring that the probe covers all exons and flanking regulatory regions (such as the promoter region) of the CDKN2A gene. The presence of homozygous deletions (loss of both alleles) is determined by comparing the sequencing depth of tumor samples with that of normal controls. A coverage depth of less than 30% of the normal value (or a log2 ratio < -1.5) is generally considered indicative of homozygous deletion.
[0075] In this embodiment, 301 cases that had undergone NGS / FISH were identified as having homozygous deletion of the CDKN2A gene. Immunohistochemical staining was performed, and then scoring was conducted.
[0076] In some embodiments, the methods for determining the first optimal cutoff threshold and the second optimal cutoff threshold based on the first ROC curve and the second ROC curve include the Youden index method, the geometric distance method, the cost-benefit ratio minimization method, and the Youden index method.
[0077] The Youden index method is suitable for scenarios where sensitivity and specificity are equally important, such as initial screening. Its core calculation formula is as follows:
[0078] J = Sensitivity + Specificity - 1
[0079] In the formula, J represents the Youden index, Sensitivity represents sensitivity, Specificity represents specificity, Sensitivity = TPR = TP / (TP+FN), Specificity = TN / (TN+FP), and False positive rate (FPR) = 1 - Specificity = FP / (TN+FP).
[0080] When J=0, TPR=FPR, and the model performance is equivalent to random prediction; when J=1, TPR=1 and FPR=0, the model achieves perfect classification (no missed diagnoses and no false diagnoses); in real-world scenarios, the J value is usually between 0 and 1, and the higher the value, the closer the model is to the ideal classifier.
[0081] Specifically, in CDKN2A gene deletion screening, the scoring data (e.g., first score, second score) are sorted, and each possible cutoff point is taken as a candidate threshold (e.g., 0.1, 0.2, ..., 0.9). The J value for each threshold is calculated. For each threshold, the corresponding TPR and specificity are calculated, and substituted into J = TPR + specificity - 1. The threshold with the largest J value is selected as the optimal cutoff threshold.
[0082] The geometric distance method is suitable for scenarios where both TPR and FPR need to be optimized simultaneously, such as diagnostic tests. The core formula of the geometric distance method is as follows:
[0083] d = (1 - Sensitivity) 2 +(1-Specificity) 2
[0084] In the formula, d represents the geometric distance. The threshold that minimizes d is found as the optimal cutoff threshold.
[0085] The cost-benefit ratio minimization method is suitable for scenarios where the cost difference between false positives and negatives is clear, such as cancer screening. Its core calculation formula is as follows:
[0086] Cost=CFPCFN×(1-Prevalence)PrevalenceCost=CFNCFP×Prevalenc e(1-Prevalence)
[0087] The Youden Index method is suitable for scenarios where sensitivity needs to be prioritized, such as infectious diseases. Its core calculation formula is as follows:
[0088] Jw=w1·Sensitivity+w2·SpecificityJw
[0089] In some embodiments, ROC curves are generated through the following steps.
[0090] Step 1: Data preparation.
[0091] Gold standard: Using the homozygous deletion of CDKN2A detected by FISH / NGS as the true label.
[0092] Predictor variables: continuous scores of MTAP expression (such as H-score or AOD optical density value).
[0093] Step 2: Generate ROC curves.
[0094] The application software IBM SPSS Statistics 26 was used to generate ROC curves, where the first ROC curve generated by the first score (manual H-score) is shown below. Figure 3 As shown, the second ROC curve generated by the second score (AOD score) is as follows: Figure 4 As shown, the first optimal cutoff threshold was determined to be 37.50, and the second optimal cutoff threshold was determined to be 16.65.
[0095] S50: Based on the first ROC curve and the second ROC curve, determine the first optimal cutoff threshold and the second optimal cutoff threshold respectively, and use the first optimal cutoff threshold and the second optimal cutoff threshold as the criteria for screening for homozygous deletion of CDKN2A, so as to achieve screening for homozygous deletion of CDKN2A.
[0096] In some embodiments, the screening of CDKN2A homozygous deletions is achieved by using a first optimal cutoff threshold and a second optimal cutoff threshold as the criteria for screening. This includes: if the first score of the sliced digital image is less than or equal to the first optimal cutoff threshold or the second score is less than or equal to the second optimal cutoff threshold, the sample corresponding to the sliced digital image is determined to be a CDKN2A homozygous deletion.
[0097] The theoretical basis for the feasibility of this application is given below, including biological basis, statistical basis and clinical application guidelines, which are explained in detail below.
[0098] Biological basis: Co-deletion mechanism of CDKN2A and MTAP.
[0099] (1) Gene location and co-deletion phenomenon
[0100] The CDKN2A gene (encoding the p16 protein) and the MTAP gene (methionine adenosine transferase) are both located in the 9p21.3 region of human chromosomes, physically adjacent (approximately 100-200 kb apart). In tumors, large deletions (homozygous deletions) in the 9p21.3 region often lead to the simultaneous loss of both the CDKN2A and MTAP genes, with a co-deletion rate reaching 90%-95% (e.g., glioma, mesothelioma, melanoma).
[0101] (2) Theoretical basis: The loss of MTAP protein expression can be used as a reliable alternative marker for homozygous loss of CDKN2A.
[0102] (3) Functional correlation: MTAP is involved in the methionine metabolic pathway. Its deficiency leads to tumor cells' dependence on exogenous purines. However, this biological process is not directly related to the CDKN2A deficiency (abnormal cell cycle regulation). The correlation between the two is only based on genomic co-deletion.
[0103] Statistical basis: ROC curve and threshold optimization.
[0104] (1) Clinical significance of ROC curve.
[0105] The area under the ROC curve (AUC) reflects the ability of the MTAP score to distinguish between CDKN2A deficiency and non-deficiency.
Claims
1. A method for detecting homozygous deletion of CDKN2A in central nervous system gliomas, characterized in that, The method comprises the following steps: obtaining a slice digital image; calculating a first score of the slice digital image; inputting the slice digital image into a pre-trained deep learning model to automatically identify and score the MATP signal in the image, and obtaining a second score; comparing the first score and the second score with the CDKN2A gene homozygous deletion gold standard of the corresponding sample respectively, and drawing a first ROC curve and a second ROC curve; determining a first optimal cutoff threshold and a second optimal cutoff threshold based on the first ROC curve and the second ROC curve respectively, taking the first optimal cutoff threshold and the second optimal cutoff threshold as the judgment basis for screening CDKN2A homozygous deletion, and realizing the screening of CDKN2A homozygous deletion.
2. The method of claim 1, wherein, The slice digital image is obtained by the following method: scanning the stained image by using a digital panoramic scanner to obtain the slice digital image; wherein the stained image is obtained by performing MTAP protein immunohistochemical staining on multiple paraffin sections of tissues from central nervous system glioma.
3. The method of claim 1, wherein, The first score of the slice digital image is calculated by the following formula: where H represents the first score, i represents the MTAP staining intensity value, I i represents the tumor area with staining intensity of cells with MTAP staining intensity value i, and N is the H-score value for MTAP staining intensity.
4. The method of claim 1, wherein, The slice digital image is input into a pre-trained deep learning model to automatically identify and score the MATP signal in the image, and the second score is calculated by the following formula: AOD = IntDen / Count In the formula, AOD represents the second score, IntDen represents the integral optical density, and Count is the pixel count.
5. The method of claim 1, wherein, The deep learning model is an image recognition model based on a deep convolutional neural network.
6. The method of claim 1, wherein, Taking the first optimal cutoff threshold and the second optimal cutoff threshold as the judgment basis for screening CDKN2A homozygous deletion, and realizing the screening of CDKN2A homozygous deletion, comprises: If the first score of the slice digital image is less than or equal to the first optimal cutoff threshold or the second score is less than or equal to the second optimal cutoff threshold, it is determined that the sample corresponding to the slice digital image is CDKN2A homozygous deletion.
7. The method of claim 1, wherein, The method of determining the first optimal cutoff threshold and the second optimal cutoff threshold based on the first ROC curve and the second ROC curve respectively comprises Youden index method, geometric distance method, cost-benefit ratio minimization method and Youden index method.
8. A detection device for screening homozygous deletion of CDKN2A in central nervous system glioma, characterized by, The method comprises the following steps: An image acquisition module configured to obtain a slice digital image; A first score calculation module configured to calculate a first score of the slice digital image; A second score calculation module configured to input the slice digital image into a pre-trained deep learning model to automatically identify and score the MATP signal in the image, and obtain a second score; An ROC curve generation module configured to compare the first score and the second score with the CDKN2A gene homozygous deletion gold standard of the corresponding sample respectively, and draw a first ROC curve and a second ROC curve; An optimal cutoff threshold determination module configured to determine a first optimal cutoff threshold and a second optimal cutoff threshold based on the first ROC curve and the second ROC curve respectively, take the first optimal cutoff threshold and the second optimal cutoff threshold as the judgment basis for screening CDKN2A homozygous deletion, and realize the screening of CDKN2A homozygous deletion.
9. An electronic device, comprising: The method comprises the following steps: a processor, and a memory connected with the processor in communication; the memory stores computer-executed instructions; the processor executes the computer-executed instructions stored in the memory to implement the method according to any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, the computer-readable storage medium stores computer-executed instructions, and the computer-executed instructions are executed by the processor to implement the method according to any one of claims 1-7.