Drug recommendation method based on diffuse large B-cell lymphoma medical image and computer model

Abstract

The invention relates to a drug recommendation method based on a diffuse large B-cell lymphoma medical image and a computer model. The drug recommendation method is based on a typing method, typing takes a tumor maximum mass diameter (MTD) as a core, and patients are divided into a non-large mass type (MTDlt, 5cm), a medium large mass type (5cm < = MTDlt, 10cm) and a huge mass type (MTD > = 10cm) in combination with CD58, STAT6 and EBF1 gene mutation probabilities, MYC, BCR, PI3K and other carcinogenic pathways and cell proliferation and sugar metabolism pathway activation degrees. The computer model adopts a deep learning architecture, can evaluate prognosis and guide treatment by inputting imaging photos, automatically analyzing the characteristics of a packet block, determining typing and outputting individualized drug recommendation in combination with the characteristics of molecules and pathways, is suitable for the medical field of hospitals, and particularly provides a convenient tool for mechanisms lacking gene sequencing conditions.

Description

Technical Field

[0001] This invention belongs to the field of computer-aided disease prediction and relates to a drug recommendation method and computer model based on medical images of diffuse large B-cell lymphoma. Background Technology

[0002] Diffuse large B-cell lymphoma (DLBCL) is a group of malignant and aggressive tumors originating from mature B cells. It is the most common type of non-Hodgkin lymphoma, accounting for 30%-40% of all non-Hodgkin lymphoma (NHL) cases. DLBCL exhibits high heterogeneity; 60%–65% of patients can achieve a cure with the first-line R-CHOP regimen, but 10%–15% develop primary resistance, and 30% relapse after complete remission. The inventor's previous patent CN114277134B proposed a DLBCL subtyping model based on 18 genes, which uses the PAM algorithm to cluster into five types, including MCD and N1. This model shows high consistency with the NEJM50 gene model, and is cost-effective for clinical application, with good kit performance. However, some parameters are not fully covered, such as missing boundary values ​​in gene testing; and some steps could be further standardized.

[0003] Currently, the academic community has devoted considerable effort to exploring the prognostic significance of clinical features, genetic alterations, and the tumor immune microenvironment in DLBCL. Among the many clinical features, large masses are a common and clinically significant characteristic of DLBCL. Large masses are significantly associated with poor prognosis in patients who are newly diagnosed, relapsed / refractory, or receiving CAR-T cell therapy. However, it is worth noting that there is currently no consensus on the optimal threshold for defining large masses, with cutoff values ​​ranging from 5.0 to 10.0 cm used in various clinical trials. For example, the MabThera international research group found a linear correlation between the largest tumor diameter and prognosis, ultimately using 10 cm as the cutoff value; another study at MD Anderson Cancer Center used a cutoff value of 5.0 cm; and the RICOVER-60 trial used 7.5 cm as the defining standard. This variability reflects the lack of standardization in the definition of large masses. Furthermore, although its prognostic significance is clear, previous studies have largely focused on clinical outcomes, and the underlying biological mechanisms of large masses in DLBCL remain poorly understood, and there are no individualized treatment strategies for patients with different tumor diameters. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the purpose of this invention is to provide a drug recommendation method and computer model based on medical images of diffuse large B-cell lymphoma. This method is based on a classification method and proposes the biological characteristics of each classification. In addition, according to the mutation characteristics of patients with masses, the signaling pathways and immune microenvironment characteristics of each mutation are analyzed, and further personalized treatment recommendations are made.

[0005] Technical Solution 1, a drug recommendation method based on medical imaging of diffuse large B-cell lymphoma, comprising the following steps: S1. Obtain medical imaging files of patients with diffuse large B-cell lymphoma who have masses; S2. Standardize the digital image file, segment the tumor mass region, and calculate the maximum mass diameter (MTD) within the tumor mass region; S3. Compare the MTD with a preset threshold to determine the classification result; S4. Based on the typing results and the image features of the digital image files, predict the mutation probability of the target gene and the degree of pathway activation. The mutation probability of the target gene includes the mutation probability of CD58, STAT6, and EBF1 genes, and the degree of pathway activation includes the activation degree of oncogenic pathways, cell proliferation pathways, and glucose metabolism pathways. S5. The classification results, target gene mutation probability, pathway activation level, and preset treatment rule base are matched to generate a structured report containing recommended drug names, medication basis, and recommendation priority.

[0006] Furthermore, using the maximum mass diameter (MTD) as the core classification criterion, patients with diffuse large B-cell lymphoma and masses are divided into the following three types: Non-large mass type: MTD < 5cm; Medium to large mass type: 5cm ≤ MTD < 10cm; Giant mass type: MTD≥10cm.

[0007] Furthermore, among the mutation probabilities of the CD58, STAT6, and EBF1 genes, the mutation probability of the above genes is low in patients with non-large mass type; the mutation probability is moderate in patients with medium-sized mass type, which is significantly higher than that in patients with non-large mass type; and the mutation probability is the highest in patients with giant mass type, which is significantly higher than that in patients with non-large mass type and medium-sized mass type.

[0008] Furthermore, patients with CD58 mutations exhibit a severely depleted immune microenvironment with minimal immune cell infiltration, dominated by malignant B cells; patients with STAT6 mutations have an immune microenvironment dominated by fibroblasts, endothelial cells, naive T cells, CD4-Treg cells, CD4-Th1-like cells, and some CD8-Teff cells; and patients with EBF1 mutations have an immune microenvironment dominated by malignant B cells, fibroblasts, endothelial cells, naive T cells, and some CD8-Teff cells.

[0009] Furthermore, the oncogenic pathways include the MYC, BCR, PI3K, JAK, and JAK2 oncogenic pathways; in patients with non-mass-type tumors, the activation levels of the MYC, BCR, PI3K, JAK, and JAK2 oncogenic pathways are low; in patients with moderately large masses, the activation levels of the MYC, BCR, PI3K, JAK, and JAK2 oncogenic pathways are moderate, significantly higher than in patients with non-mass-type tumors; in patients with giant masses, the activation levels of the MYC, BCR, PI3K, JAK, and JAK2 oncogenic pathways are the highest, significantly higher than in patients with non-mass-type tumors and moderately large masses.

[0010] Furthermore, in patients with non-large masses, the activation of cell proliferation and glucose metabolism pathways was weaker; in patients with medium-sized masses, the activation levels of cell proliferation and glucose metabolism pathways were moderate; and in patients with giant masses, the activation of cell proliferation and glucose metabolism pathways was strongest.

[0011] Furthermore, the above method was established based on the mass diameter data and prognostic analysis results of 939 DLBCL patients. The maximum mass diameter of the tumor can be obtained through imaging examinations such as PET-CT, CT or color Doppler ultrasound to complete the classification, which can be used to assess the patient's prognosis and guide the formulation of individualized treatment plans.

[0012] Furthermore, patients with non-large mass type have a relatively good prognosis, with high progression-free survival (PFS) and overall survival (OS); patients with medium-sized mass type have a prognosis between non-large mass type and giant mass type, with significantly lower progression-free survival and overall survival than non-large mass type, and statistically different from giant mass type; patients with giant mass type have the worst prognosis, with significantly lower progression-free survival and overall survival than the first two types.

[0013] Technical Solution 2, a computer model constructed based on the aforementioned drug recommendation method for diffuse large B-cell lymphoma medical images, wherein the computer model adopts a multi-module serial deep learning architecture, including: Image analysis module: Built on a convolutional neural network, it is used to extract tumor mass features from imaging images and automatically calculate the maximum mass diameter (MTD). The classification matching module compares the MTD output by the image analysis module with the preset classification threshold to determine the patient's classification. Feature prediction module: Based on recurrent neural network or Transformer architecture, it combines the correlation data of imaging features and molecular and pathway features in the training set to predict the probability of CD58 / STAT6 / EBF1 gene mutation and the degree of pathway activation in patients. Drug recommendation module: Embedded with a preset treatment rule library, it outputs a matching personalized drug regimen based on the subtyping results and predicted molecular and pathway characteristics.

[0014] Furthermore, the rule base of the drug recommendation module is constructed based on the molecular characteristics, pathway characteristics, and clinical efficacy data of each subtype, and its contents are as follows: (1) Non-large mass type (MTD<5cm) For patients without CD58 / STAT6 mutations: standard treatment regimens (such as R-CHOP) are recommended, with the addition of TIGIT inhibitors or CTLA-4 inhibitors in case of disease relapse. For CD58 mutation: conventional treatment combined with PD-L1 inhibitors or IDO1 inhibitors is recommended; For patients with STAT6 mutations: it is recommended to combine standard treatment with JAK2 inhibitors, or to add TIGIT / CTLA-4 inhibitors based on the expression of immune checkpoint molecules. (2) Medium-sized large lumps (5cm≤MTD<10cm) For patients without CD58 / STAT6 mutations: conventional treatment combined with TIGIT / CTLA-4 inhibitors, CD70 inhibitors, 4-1BB agonists, or BCR / PI3K / JAK pathway inhibitors is recommended. CD58 mutation present: Standard treatment plus PD-L1 / IDO1 inhibitors are recommended; STAT6 mutation present: Standard treatment plus JAK2 inhibitor recommended; (3) Giant mass type (MTD≥10cm) For patients without CD58 / STAT6 mutations: conventional treatment combined with CD70 inhibitors, 4-1BB agonists, or BCR / PI3K / JAK pathway inhibitors is recommended. CD58 mutation present: Standard treatment plus PD-L1 / IDO1 inhibitors are recommended; STAT6 mutation present: Standard treatment plus JAK2 inhibitor recommended.

[0015] Technical solution 3, an electronic device, including a memory and a processor, wherein the processor is used to execute a program in the memory to implement the drug recommendation method based on medical images of diffuse large B-cell lymphoma with masses.

[0016] Technical solution 4, a storage medium containing computer-executable instructions, which, when executed by a computer processor, is used to execute the aforementioned drug recommendation method based on medical images of diffuse large B-cell lymphoma with masses.

[0017] Technical Solution 5: A biomarker recommended for patients with diffuse large B-cell lymphoma and masses, wherein the biomarker includes CD58, EBF1, and STAT6 genes, and the mutation frequency of these genes is positively correlated with the diameter of the mass.

[0018] Furthermore, the CD58 mutation corresponds to an immune-depleted microenvironment and is associated with drug sensitivity to PD-L1 and IDO1 inhibitors. Computer models can automatically match these immunotherapies by predicting this mutation status. The STAT6 mutation is associated with JAK2 pathway activation and is significantly correlated with treatment response to JAK2 inhibitors. The model can prioritize recommending these targeted drugs based on mutation prediction results. The EBF1 mutation affects the B cell differentiation pathway and has a significant synergistic effect with BCR / PI3K pathway inhibitors. The model will use this as a key reference indicator for combination therapy.

[0019] Furthermore, combining mass diameter classification and biomarker status, the classification is refined into the following subtypes, each corresponding to a pre-defined drug recommendation logic: Low-risk molecular subtype type I: MTD < 5cm + no mutations in CD58 / EBF1 / STAT6. The model automatically recommends a basic regimen (such as R-CHOP) and marks it as "no need for intensive treatment" because this subtype has a high response rate to conventional chemotherapy.

[0020] Intermediate-risk molecular subtype II: 5cm≤MTD<10cm + any gene mutation (not all mutations). The model recommends a combination regimen based on the mutation type: if STAT6 mutation, R-CHOP + JAK2 inhibitor is recommended; if CD58 mutation, R-CHOP + PD-L1 inhibitor is recommended, and the priority is set to "intermediate".

[0021] High-risk molecular subtype III: MTD ≥ 10 cm + at least two gene mutations, model-triggered intensive treatment recommendation, such as R-CHOP + BCR inhibitor + PD-L1 inhibitor, with the note "close monitoring of efficacy required".

[0022] Compared with the prior art, the technical solution of the present invention has the following improvements and significant advantages: 1. Establish a standardized classification system: This invention is the first to propose a three-tiered standard method for DLBCL patients with masses, based on the maximum tumor diameter (MTD). It clearly distinguishes between non-large masses (MTD<5cm), medium-sized masses (5cm≤MTD<10cm), and giant masses (MTD≥10cm), and achieves accurate assessment of the prognosis (overall survival) of patients in each subtype, providing a quantitative basis for the development of personalized treatment plans.

[0023] 2. Revealing the biological characteristics of typing: This invention systematically elucidates the molecular characteristics of different subtypes (the mutation frequency of CD58, EBF1, and STAT6 genes increases with the diameter of the mass), signaling pathway characteristics (the activation degree of oncogenic pathways such as MYC and BCR, as well as proliferation / glucose metabolism pathways, is positively correlated with the size of the mass), and immune microenvironment characteristics (such as CD58 mutation patients exhibiting a state of severe immune exhaustion). This provides clear biological markers for targeted therapy and promotes the transformation of personalized medicine from empirical medicine to precision medicine.

[0024] 3. Lower the threshold for clinical application: This invention overcomes the limitations of high cost and long time consumption in RNA sequencing. It can predict pathway activation and immune microenvironment characteristics by using the diameter of the mass and mutation information, without relying on complex detection technologies. This greatly reduces the difficulty of implementing personalized treatment and makes it convenient for primary healthcare institutions to use.

[0025] 4. Improve the feasibility of clinical application: This invention obtains classification data based on conventional imaging examinations such as PET-CT, CT, or color Doppler ultrasound. It is easy to operate, widely applicable, and can be quickly implemented in hospitals at all levels. It helps doctors to accurately guide medication, significantly improves the survival rate and quality of life of patients with masses, and has broad prospects for clinical application.

[0026] 5. The computer model provided by this invention has three major advantages: (1) High degree of automation, automatically completes MTD measurement, typing and drug recommendation, reducing human error; (2) High degree of personalization and precision, combining molecular and pathway characteristics to achieve "tailor-made" treatment, avoiding "one-size-fits-all"; (3) Strong scalability, supports the inclusion of new clinical data for continuous iteration and optimization, especially providing efficient medication guidance tools for institutions lacking gene sequencing or pathway analysis conditions, effectively assisting doctors in developing individualized plans and improving patient prognosis. Attached Figure Description

[0027] Figure 1 Comparison of prognostic survival curves for different mass diameters in DLBCL; Figure 1 A reflects the difference in disease progression-free survival among patients in different MTD groups. Figure 1 B reflects overall survival differences. Figure 2A comparison chart of gene mutation frequencies in patients with different mass diameters in DLBCL. Figure 3 Comparison of the degree of activation of oncogenic pathways in patients with different mass diameters of DLBCL; Figure 4 Comparison of immune cell infiltration scores for patients with different mass diameters in DLBCL; Figure 5 Comparison of mRNA expression of immune checkpoint molecules for different DLBCL mass diameter subtypes; Figure 6 Heatmaps showing the characteristics of single-cell immune cells in patients with different gene mutations in DLBCL. Detailed Implementation

[0028] To enable those skilled in the art to better understand the technical solutions of the present invention, the present invention will be described in detail below with reference to specific embodiments. It should be noted that the following embodiments will help those skilled in the art to further understand the present invention, but do not limit the present invention in any way. It should be pointed out that those skilled in the art can make several modifications and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0029] All raw materials used in this invention are not particularly limited in their source; they can be purchased from the market or prepared using conventional methods known to those skilled in the art.

[0030] Example 1 (1) Data acquisition: Mass diameter data were obtained from 939 patients with DLBCL, based on the literature ( Lancet Oncol. 12, 1013–1022 (2011); J. Clin. Oncol. 38, 3003–3011 (2020); J. Clin. Oncol. 26, 2258–2263 (2008); J. Clin. Oncol. 28, 4170–4176 (2010); Lancet Oncol. The data commonly used to define the diameter of large blobs, as mentioned in 9, 105–116 (2008), are grouped into four groups: The first group had masses with a diameter less than 5 cm; the second group had masses with a diameter greater than or equal to 5 cm but less than 7.5 cm; the third group had masses with a diameter greater than or equal to 7.5 cm but less than 10 cm; and the fourth group had masses with a diameter greater than or equal to 10 cm. Prognostic analysis of these four groups revealed no statistically significant differences in prognosis or remission rates between the second and third groups. However, statistically significant differences were found between the first and second groups, and between the third and fourth groups. Based on this prognostic characteristic, masses with diameters of 5-7.5 cm and 7.5-10 cm were grouped together for prognostic analysis, revealing differences in prognosis among the three groups. Therefore, masses with a diameter less than 5 cm were defined as non-large masses, masses with a diameter greater than or equal to 5 cm but less than 10 cm as medium-sized masses, and masses with a diameter greater than or equal to 10 cm as large masses.

[0031] Experimental results are as follows Figure 1 As shown, patients with DLBCL masses were divided into three groups based on their diameter, and there were statistically significant differences in prognosis (progression-free survival, PFS, and overall survival, OS) among the three groups.

[0032] Figure 1 A and Figure 1 Figure B shows the survival curves for prognostic analysis of DLBCL patients based on mass diameter classification (N=673). Patients were clearly divided into three groups: MTD < 5cm, 5cm ≤ MTD < 10cm, and MTD ≥ 10cm (MTD, maximum tumor diameter). Both figures show that as the mass diameter increases, the survival curves (progression-free survival, PFS, and overall survival, OS) gradually decrease, with a p-value < 0.001 between groups, indicating significant prognostic differences. Mass size is closely related to the prognosis of DLBCL patients; the larger the diameter, the worse the prognosis. This validates the rationality of the above-mentioned integrated grouping and three-level classification based on prognostic characteristics, clarifying that it can serve as an effective tool for prognostic assessment, providing a reference for treatment selection, and supporting the clinical value of mass classification in prognostic evaluation.

[0033] (2) Mutation characteristics: Mutation characteristic analysis of the three types of masses revealed that mutations in the CD58, STAT6, and EBF1 genes were positively correlated with the diameter of the masses.

[0034] like Figure 2 As shown, DNA sequencing was performed on the patient, and genomic DNA was extracted from either frozen tumor tissue (using the QIAampDNA Mini Kit, Qiagen, Germany) or formalin-fixed paraffin-embedded (FFPE) tumor tissue (using the GeneRead DNAFFPE Tissue Kit, Qiagen), with the procedures strictly following the manufacturer's guidelines.

[0035] To establish criteria for identifying somatic mutations and exclude interference from germline polymorphisms, 42 paired peripheral blood samples were randomly selected for WES and WGS testing. Targeted sequencing data were analyzed using the Haplotype Caller and GATK Unified Genotyper tools in the GATK v3.7.0 genome analysis toolkit for single nucleotide variant (SNV) and indel detection, and genome localization was performed using the UCSC Genome Explorer (http: / / genome.ucsc.edu). The reference genome was the Refseq database (human reference genome version hg19), and detected SNVs and indels were filtered using a self-built workflow based on the aforementioned software.

[0036] Subsequently, 55 commonly mutated genes were compared among the three groups based on the mass diameter. It was found that the mutation rates of CD58, STAT6, and EBF1 increased with increasing mass diameter. The figure shows a bar chart comparing gene mutation frequencies for the three mass types (MTD<5cm, 5cm≤MTD<10cm, and MTD≥10cm), with the horizontal axis representing different genes and the vertical axis representing mutation frequency. It is evident that different genes exhibit differentiated mutation distributions under different mass subtypes. The mutation frequency of some genes (such as CD58, STAT6, and EBF1) in the MTD≥10cm group showed a significantly higher trend compared to the MTD<5cm and 5cm≤MTD<10cm groups, initially demonstrating a correlation between gene mutations and mass diameter. The data distribution shows that CD58, STAT6, and EBF1 genes have higher mutation frequencies in larger mass subtypes (MTD≥10cm), indicating that these three types of gene alterations may be enriched as the mass increases in size. This provides evidence at the mutation frequency level to support further research on the mechanisms by which genes affect mass progression, prognosis, and the selection of therapeutic drugs.

[0037] (3) Characteristics of carcinogenic pathways: Oncogenic pathway analysis of the three types of masses revealed that the MYC, BCR, PI3K, JAK, and JAK2 pathways gradually increased with increasing mass diameter. Furthermore, cell proliferation and glucose metabolism pathways also increased with increasing mass diameter.

[0038] like Figure 3As shown, double-stranded cDNA was synthesized using the VAHTS mRNA-seq V3 library construction kit (Nanjing Novozymes, China). The purified cDNA underwent pre-amplification to construct a primary library. Qualified DNA libraries were enriched with Dynabeads MyOneStreptavidin T1 magnetic beads and then amplified to obtain the final library. After quality control, paired-end 150bp high-throughput RNA sequencing was performed using the Illumina NovaSeq sequencing platform (Illumina, USA). The paired-end RNA sequencing data was pseudo-aligned to the GRCh38 transcriptome reference sequence (Ensembl database release-106 version) using Kallisto software (v0.46.0) with default parameters. Gene expression levels are expressed as raw counts and TPMs. The raw Kallisto output results were integrated using the R package "tximport" (v1.24.0), and the sum of the counts and TPM values ​​of all alternatively spliced ​​transcripts of the same gene was used as the expression level of that gene. Protein-coding genes, immunoglobulin (IG-C / V / D / J) genes, and T-cell receptor (TR-C / V / D / J) genes were retained, while histone-related genes and mitochondrial genes were removed, ultimately retaining approximately 20,000 genes for downstream analysis. Before performing LME classification and calculating gene expression characteristics, the log2-transformed TPM values ​​(log2TPM) were standardized using the R package "preprocessCore" (v1.58.0). PCA analysis was performed using the R package "limma" (v3.48.3) to confirm the elimination of batch effects caused by sequencing time. Based on the current RNA sequencing data, the corresponding gene expression matrix was standardized using GSVA analysis. Then, based on the expression level of each sample in the gene expression matrix, its score relative to different oncogenic pathway sets was calculated. Oncogenic pathway scores were analyzed for the three types of masses. With increasing mass diameter, the MYC, BCR, PI3K, JAK, and JAK2 pathways gradually increased. In addition, cell proliferation and glucose metabolism pathways also increased with increasing mass diameter. This figure is a box plot showing the activation levels of oncogenic pathways in three types of masses (MTD<5cm, 5cm≤MTD<10cm, MTD≥10cm). The vertical axis represents the standardized score of pathway activation, and the horizontal axis represents different pathways (JAK, JAK2, etc.). It can be seen that as the mass diameter increases (from MTD<5cm to MTD≥10cm), the activation scores of pathways such as MYC, BCR, and JAK generally show an upward trend. For example, the scores of MYC and BCR pathways in the MTD≥10cm group are significantly higher than those in the other two groups, demonstrating the correlation between pathway activation and mass diameter.

[0039] Data show that as the mass diameter increases, the activation levels of oncogenic pathways such as MYC, BCR, and JAK, as well as cell proliferation and glucose metabolism pathways, increase. This indicates a positive correlation between oncogenic, proliferation, and glucose metabolism pathways and mass diameter, suggesting that abnormal activation of these pathways may play a role in the mass enlargement process. This provides pathway-level evidence for elucidating the molecular mechanisms of DLBCL mass progression.

[0040] (4) Characteristics of the immune microenvironment: Immune microenvironment analysis of type III masses revealed that CD4 levels increased with increasing mass diameter. + T cells and CD8 + T cells and cDC cells gradually decreased, exhibiting characteristics of immune exhaustion. In addition, mast cells and pro B cells increased with increasing diameter. Analysis of immune checkpoint molecule expression levels revealed that CTLA-4, TIGIT, ICOS, and CD28 expression decreased with increasing mass diameter, while CD70 and 4-1BBL expression increased with increasing mass diameter.

[0041] like Figure 4 As shown, based on the RNA sequencing data obtained above, using the xCELL website and the ssGSEA algorithm, the gene expression profile of each sample was transformed into a cross-sample representation of the abundance scores of immune and stromal cell types by comparing the gene expression data of each sample with the immune cell gene set. Specifically, ssGSEA first sorts all genes according to their expression levels from highest to lowest and calculates the cumulative distribution function of genes with higher expression levels within a given gene set. This cumulative distribution function is called the gene set enrichment score (GSE). Then, for each sample, the expression levels of all genes in that sample are arranged in descending order, and the gene set enrichment score corresponding to each position is calculated. Finally, the scores at these positions are averaged or weighted to obtain the ssGSEA score of that sample on that gene set, which is used to estimate the relative abundance of that immune cell type in that sample. Immune microenvironment analysis through comparison of the three types of mass fractions revealed that as the mass diameter increases, CD4... + T cells and CD8 + T cells and cDC cells gradually decrease, exhibiting characteristics of immune exhaustion. In addition, mast cells and proB cells increase with increasing diameter. This figure shows the immune cell infiltration in the three types of masses (MTD<5cm, 5cm≤MTD<10cm, MTD≥10cm). The vertical axis represents the immune cell infiltration score, and the horizontal axis represents various types of immune cells (CD4+, CD5+, CD6+, CD4+, CD6 ... + T, CD8 + T, etc. It can be seen that as the mass diameter increases (from MTD<5cm to MTD≥10cm), CD4...+ T, CD8 + The overall infiltration scores of T cells and other cells showed a decreasing trend, while the scores of mast cells and other cells showed an increasing trend, visually demonstrating the change in immune cell composition with the diameter of the mass. CD4 + T, CD8 + T cells and cDC cells decrease with increasing diameter, consistent with the trend of "immune exhaustion"; mast cells and other cells increase. These dynamic changes in immune cells provide a basis for understanding the evolution of immune function during the progression of DLBCL masses.

[0042] like Figure 5 As shown, analysis of immune checkpoint molecule expression revealed that the expression of CTLA-4, TIGIT, ICOS, and CD28 decreased with increasing mass diameter, while the expression of CD70 and 4-1BBL increased with increasing mass diameter. This figure shows the mRNA expression results of immune checkpoint molecules (ICOS, TIGIT, etc.) under different mass subtypes (MTD<5cm, 5cm≤MTD<10cm, MTD≥10cm). The vertical axis represents the standardized mRNA expression level, and the horizontal axis represents the diameter group. It can be seen that with increasing mass diameter (e.g., MTD≥10cm group), the expression of molecules such as ICOS and CTLA4 decreased, while the expression of CD70 and 4-1BBL increased, demonstrating a correlation between immune checkpoint molecule expression and mass diameter.

[0043] (5) Analysis of the immune microenvironment characteristics of patients with mutations: Compared to the control group, patients with CD58 mutations exhibited a severely immune-depleted microenvironment with minimal immune cell infiltration and a predominance of malignant B cells. Patients with STAT6 mutations, compared to the control group, showed a predominance of fibroblasts, endothelial cells, naive T cells, CD4-Tregs, CD4-Th1-like cells, and some CD8-Teff cells. Patients with EBF1 mutations, compared to the control group, showed a predominance of malignant B cells, fibroblasts, endothelial cells, naive T cells, and some CD8-Teff cells. These findings highlight the heterogeneity of the tumor immune microenvironment driven by specific gene alterations. Different mutations may affect the immune infiltration and matrix composition of DLBCL.

[0044] like Figure 6As shown, tumor samples were digested in GEXSCOPE tissue preservation solution (Singleron Biotechnologies, USA) at 37°C for 15 minutes, followed by separation of cells and debris using a 40μm sterile cell sieve (Corning, USA). To remove red blood cells, 2 ml of GEXSCOPE red blood cell lysis buffer (Singleron Biotechnologies, USA) was mixed with the cell suspension, incubated at 25°C for 10 minutes, centrifuged at 500×g for 5 minutes, and the pellet was resuspended in PBS buffer. Cell counting was performed using a TC20 automated cell counter (Bio-Rad, USA). Single-cell RNA libraries were constructed using the 10×Chromium single-cell platform (Chromium Single Cell 3' Library Kit, Gel Beads and Chip Kit, 10×Genomics, USA). After flow cytometry sorting, 8,000-16,500 viable hCD45+ cells (viability >80%) were loaded into each channel, with a target capture of 2,000-10,000 single cells. After forming gel microbead emulsions (GEMs), reverse transcription and amplification reactions were performed, followed by cDNA fragmentation after purification. The final library was sequenced using a standard NovaSeq 6000 platform (Illumina, USA) or MGISEQ-2000 platform (BGI Genomics, China). Data splitting was performed using Cell Ranger software (v6.0.1) provided by 10×Genomics, and reads were mapped to the GRCh38 reference genome using the STAR alignment tool to generate a feature barcode-UMI molecular counting matrix. The UMI matrix was quality controlled using the Seurat R package (v5.1.0): cells with <200 or >6,000 genes were excluded; low-quality cells with >25% mitochondrial gene expression were removed; and doublets were predicted and filtered using the DoubletFinder R package (v2.0.3). After quality control, 25,746 single cells were retained for subsequent analysis. The quality-controlled feature barcode matrix was standardized using the Seurat R package (v5.1.0): the NormalizeData function was used for library size standardization, 3,000 hypervariable genes were screened using FindVariableFeatures, and PCA analysis was performed after centering and scaling using the ScaleData function. To eliminate batch effects, the Harmony algorithm (R package v1.0) was used to correct the 30 principal components. The RunUMAP function was used for dimensionality reduction visualization, and unsupervised clustering based on Euclidean distance in the PCA space was performed using the Louvain algorithm (FindClusters function) for single-cell feature analysis of each group.This is a heatmap showing the association between cell types and gene mutations (EBF1, STAT6, CD58 mutations, and WT without mutations) in the immune microenvironment of single cells in DLBCL (diffuse large B-cell lymphoma). Cell types include various immune-related cells such as non-malignant B cells, malignant B cells, fibroblasts, endothelial cells (ECs), and naive T cells. Gene mutation types include EBF1 mutations, STAT6 mutations, CD58 mutations, and WT (without the above gene mutations, i.e., control group). The heatmap colors represent the relative proportions or associations of different cell types in the immune microenvironment under different gene mutation states (red-blue indicates a trend from high to low), demonstrating the impact of different mutations on immune cell infiltration and matrix composition. Compared to the control group, patients with CD58 mutations exhibit a severely immune-depleted microenvironment with very little immune cell infiltration and a predominance of malignant B cells. Compared with the control group, patients with STAT6 mutations mainly consisted of fibroblasts, endothelial cells, naïve T cells, CD4-Treg cells, CD4-Th1-like cells, and some CD8-Teff cells. Compared with the control group, patients with EBF1 mutations mainly consisted of malignant B cells, fibroblasts, endothelial cells, naïve T cells, and some CD8-Teff cells.

[0045] Example 2 Drug Recommendation Patients with CD58 mutations: PD-L1 or IDO1 inhibitors; Patients with STAT6 mutations: JAK / JAK2 inhibitors For non-large to moderately large masses: CTLA-4 inhibitors and TIGIT inhibitors.

[0046] For patients with moderate / large masses: BCR pathway inhibitors, PI3K pathway inhibitors, drugs targeting 4-1BB or CAR T therapy targeting 4-1BB, CD70 inhibitors or CAR NK therapy targeting CD70.

[0047] Since the activation level of the BCR and PI3K pathways in patients is positively correlated with the increase in mass diameter, the above-mentioned inhibitors can be selected for non-large / medium-sized masses. Since the mRNA expression levels of 4-1BBL and CD70 are positively correlated with the mass diameter in immune checkpoint analysis, corresponding targeted drugs or CAR T therapy and CAR NK therapy targeting the above-mentioned targets can be selected.

[0048] Patients with CD58-mutated DLBCL exhibit a severely immune-depleted microenvironment, where the surveillance and killing functions of immune cells against tumors are greatly suppressed. PD-L1 inhibitors specifically block the PD-1 / PD-L1 immunosuppressive pathway. Under normal circumstances, PD-L1 on the surface of tumor cells binds to PD-1 on immune cells, transmitting immunosuppressive signals. After using PD-L1 inhibitors, this binding is blocked, allowing immune cells to regain activity and enhance their killing effect on tumor cells. IDO1 inhibitors, by inhibiting indoleamine 2,3-dioxygenase 1, reduce tryptophan metabolic consumption, break tumor-induced immune tolerance, thereby reducing immunosuppression and helping the immune system fight tumors.

[0049] STAT6 is a key pivot molecule in the JAK / STAT signaling pathway, playing a crucial role in cell proliferation, differentiation, and immune regulation. Mutations in STAT6 can lead to functional and phenotypic alterations in related cells such as fibroblasts and endothelial cells, thereby affecting the tumor microenvironment. JAK / JAK2 inhibitors specifically inhibit JAK kinase activity, blocking the overactivation of the JAK / STAT signaling pathway. This can correct the abnormal cell proliferation and immune regulatory imbalance caused by STAT6 mutations, interfering with tumor cell survival and development, while simultaneously reshaping the tumor microenvironment to make it more immune-friendly.

[0050] In the immune microenvironment of DLBCL, changes in immune cells and immune checkpoint molecules affect immune function. CTLA-4 inhibitors, by blocking CTLA-4-mediated immunosuppressive signals, can enhance T cell activation and proliferation, allowing the immune system to better recognize and attack tumor cells. TIGIT inhibitors target the immune checkpoint protein TIGIT, which inhibits the activity of NK cells and T cells. Using TIGIT inhibitors can relieve this inhibition, restoring the immune function of NK cells and T cells, especially in patients with non-large / medium-sized tumors, enhancing the body's immune surveillance and killing capabilities against tumors from an immunomodulatory perspective.

[0051] Example 3 This embodiment verifies the effectiveness of the solution provided in embodiment 2 based on the data provided in embodiment 1 above. The steps are as follows: (1) After the patient is diagnosed with DLBCL, a PET-CT examination is performed. Based on the image results, it is determined whether the patient has a mass and the diameter of the mass. At the same time, gene mutation detection is performed. (2) Based on the above classification results and test results, formulate a treatment plan: 1) Non-large mass type with a diameter of <5cm No CD58 / STAT6 mutation: Standard treatment regimen is used, with TIGIT or CTLA-4 inhibitors added when the disease relapses.

[0052] CD58 mutation present: Add PD-L1 or IDO1 inhibitors to the standard treatment regimen.

[0053] For patients with STAT6 mutations: add a JAK2 inhibitor to the standard treatment regimen; or add a TIGIT / CTLA-4 inhibitor based on the expression of immune checkpoint molecules.

[0054] 2) Medium-sized lumps with a diameter of 5-10cm For patients without CD58 / STAT6 mutations: conventional treatment combined with TIGIT / CTLA-4 inhibitors, or CD70 inhibitors, 4-1BB agonists, or BCR / PI3K / JAK pathway inhibitors.

[0055] CD58 mutation present: standard treatment + PD-L1 / IDO1 inhibitor.

[0056] STAT6 mutation present: standard treatment + JAK2 inhibitor.

[0057] 3) Large lumps with a diameter ≥ 10cm For patients without CD58 / STAT6 mutations: standard treatment combined with CD70 inhibitors, 4-1BB agonists, or BCR / PI3K / JAK pathway inhibitors.

[0058] CD58 mutation present: standard treatment + PD-L1 / IDO1 inhibitor.

[0059] STAT6 mutation present: standard treatment + JAK2 inhibitor.

[0060] The above description of the embodiments is provided to enable those skilled in the art to understand and use the invention. It will be apparent to those skilled in the art that various modifications can be made to these embodiments, and the general principles described herein can be applied to other embodiments without inventive effort. Therefore, the present invention is not limited to the above embodiments, and any improvements and modifications made by those skilled in the art based on the disclosure of the present invention without departing from the scope of the invention should be within the protection scope of the present invention.

Claims

1. A method for medical image-based drug recommendation for diffuse large B-cell lymphoma, characterized by, The method comprises the following steps: S1, obtaining a medical image file of a patient with diffuse large B-cell lymphoma; S2, standardizing the digital image file, segmenting the tumor mass area, and calculating the maximum tumor mass diameter MTD in the tumor mass area; S3, comparing the MTD with a preset threshold to determine the classification result; S4, based on the classification result and the image features of the digital image file, predicting the target gene mutation probability and the activation degree of the pathway, the target gene mutation probability including the mutation probability of CD58, STAT6, and EBF1 genes, and the activation degree of the pathway including the activation degree of the oncogenic pathway, the cell proliferation pathway, and the sugar metabolism pathway; S5, matching the classification result, the target gene mutation probability, the activation degree of the pathway, and the preset treatment rule library to generate a structured report containing recommended drug names, medication basis, and recommended priority.

2. The drug recommendation method based on medical imaging of diffuse large B-cell lymphoma according to claim 1, characterized in that, The maximum tumor mass diameter MTD is used as the core classification basis to classify the patients with diffuse large B-cell lymphoma with a mass into the following three types: Non-large mass type: MTD<5cm; Medium-large mass type: 5cm≤MTD<10cm; Giant mass type: MTD≥10cm. 3.The method of claim 1, wherein the method is characterized by, Among the mutation probabilities of CD58, STAT6, and EBF1 genes, the mutation probability of the above genes in non-large mass type patients is low; in medium-large mass type patients, the mutation probability is medium, which is significantly higher than that in non-large mass type; in giant mass type patients, the mutation probability is the highest, which is significantly higher than that in non-large mass type and medium-large mass type. 4.The method of claim 1, wherein the method is based on medical images of diffuse large B-cell lymphoma. The oncogenic pathways include MYC, BCR, PI3K, JAK, and JAK2 oncogenic pathways; in non-large mass type patients, the activation degree of MYC, BCR, PI3K, JAK, and JAK2 oncogenic pathways is low; in medium-large mass type patients, the activation degree of MYC, BCR, PI3K, JAK, and JAK2 oncogenic pathways is medium, which is significantly higher than that in non-large mass type; in giant mass type patients, the activation degree of MYC, BCR, PI3K, JAK, and JAK2 oncogenic pathways is the highest, which is significantly higher than that in non-large mass type and medium-large mass type.

5. The method of claim 1, wherein the method is based on medical images of diffuse large B-cell lymphoma. In non-large mass type patients, the activation of cell proliferation pathway and sugar metabolism pathway is weak; in medium-large mass type patients, the activation level of cell proliferation pathway and sugar metabolism pathway is medium; in giant mass type patients, the activation of cell proliferation pathway and sugar metabolism pathway is the strongest.

6. A computer model constructed based on the method of claim 1-6 for diffuse large B-cell lymphoma medical image-based drug recommendation, characterized in that, The computer model adopts a deep learning architecture with multiple modules in series, including: An image analysis module based on a convolutional neural network for extracting tumor mass features from an image and automatically calculating the maximum tumor mass diameter MTD; A classification matching module for comparing the MTD output by the image analysis module with a preset classification threshold to determine the classification of the patient; A feature prediction module based on a recurrent neural network or Transformer architecture to predict the mutation probability of CD58 / STAT6 / EBF1 genes and the activation degree of the pathway based on the association data between the image features and molecular and pathway features in the training set. Drug recommendation module: embed pre-set treatment rule base, according to the typing results, predicted molecular and path characteristics, output matching individualized drug scheme.

7. A computer model according to claim 6, wherein, The rule base of the drug recommendation module is constructed based on the molecular characteristics, path characteristics and clinical efficacy data of each typing, and the content is as follows: (1) Non-large mass type: No CD58 / STAT6 mutation: recommend conventional treatment, add TIGIT inhibitor or CTLA-4 inhibitor when the disease relapses; CD58 mutation exists: recommend conventional treatment combined with PD-L1 inhibitor or IDO1 inhibitor; STAT6 mutation exists: recommend conventional treatment combined with JAK2 inhibitor, or add TIGIT / CTLA-4 inhibitor according to immune checkpoint molecule expression; (2) Medium-large mass type: No CD58 / STAT6 mutation: recommend conventional treatment combined with TIGIT / CTLA-4 inhibitor, CD70 inhibitor, 4-1BB agonist, or BCR / PI3K / JAK pathway inhibitor; CD58 mutation exists: recommend conventional treatment + PD-L1 / IDO1 inhibitor; STAT6 mutation exists: recommend conventional treatment + JAK2 inhibitor; (3) Large mass type: No CD58 / STAT6 mutation: recommend conventional treatment combined with CD70 inhibitor, 4-1BB agonist, or BCR / PI3K / JAK pathway inhibitor; CD58 mutation exists: recommend conventional treatment + PD-L1 / IDO1 inhibitor; STAT6 mutation exists: recommend conventional treatment + JAK2 inhibitor.

8. An electronic device comprising a memory, a processor, characterized in that, The processor is used to execute the program in the memory, so as to realize the drug recommendation method based on diffuse large B-cell lymphoma with mass medical image as claimed in any one of claims 1-5.

9. A storage medium containing computer-executable instructions, wherein: The storage medium of the computer executable instructions, when executed by the computer processor, is used to execute the drug recommendation method based on diffuse large B-cell lymphoma with mass medical image as claimed in any one of claims 1-5.

10. A biomarker for drug recommendation for a patient with bulky disease diffuse large B-cell lymphoma, characterized in that, The biomarker includes CD58, EBF1, STAT6 gene.

Images