One in every eight men in the US is diagnosed with prostate cancer,making it the most common cancer in men.Gleason grading is one of the most essential diagnostic and prognostic factors for planning the treatment of p...One in every eight men in the US is diagnosed with prostate cancer,making it the most common cancer in men.Gleason grading is one of the most essential diagnostic and prognostic factors for planning the treatment of prostate cancer patients.Traditionally,urological pathologists perform the grading by scoring the morphological pattern,known as the Gleason pattern,in histopathology images.However,thismanual grading is highly subjective,suffers intra-and inter-pathologist variability and lacks reproducibility.An automated grading system could be more efficient,with no subjectivity and higher accuracy and reproducibility.Automated methods presented previously failed to achieve sufficient accuracy,lacked reproducibility and depended on high-resolution images such as 40×.This paper proposes an automated Gleason grading method,ProGENET,to accurately predict the grade using low-resolution images such as 10×.This method first divides the patient’s histopathology whole slide image(WSI)into patches.Then,it detects artifacts and tissue-less regions and predicts the patch-wise grade using an ensemble network of CNN and transformer models.The proposed method adapted the International Society of Urological Pathology(ISUP)grading system and achieved 90.8%accuracy in classifying the patches into healthy and Gleason grades 1 through 5 using 10×WSI,outperforming the state-of-the-art accuracy by 27%.Finally,the patient’s grade was determined by combining the patch-wise results.The method was also demonstrated for 4−class grading and binary classification of prostate cancer,achieving 93.0%and 99.6%accuracy,respectively.The reproducibility was over 90%.Since the proposedmethod determined the grades with higher accuracy and reproducibility using low-resolution images,it is more reliable and effective than existing methods and can potentially improve subsequent therapy decisions.展开更多
基金supported by Princess Nourah bint Abdulrahman University Researchers Supporting Project number(PNURSP2025R104),Princess Nourah bint Abdulrahman University,Riyadh,Saudi Arabia.
文摘One in every eight men in the US is diagnosed with prostate cancer,making it the most common cancer in men.Gleason grading is one of the most essential diagnostic and prognostic factors for planning the treatment of prostate cancer patients.Traditionally,urological pathologists perform the grading by scoring the morphological pattern,known as the Gleason pattern,in histopathology images.However,thismanual grading is highly subjective,suffers intra-and inter-pathologist variability and lacks reproducibility.An automated grading system could be more efficient,with no subjectivity and higher accuracy and reproducibility.Automated methods presented previously failed to achieve sufficient accuracy,lacked reproducibility and depended on high-resolution images such as 40×.This paper proposes an automated Gleason grading method,ProGENET,to accurately predict the grade using low-resolution images such as 10×.This method first divides the patient’s histopathology whole slide image(WSI)into patches.Then,it detects artifacts and tissue-less regions and predicts the patch-wise grade using an ensemble network of CNN and transformer models.The proposed method adapted the International Society of Urological Pathology(ISUP)grading system and achieved 90.8%accuracy in classifying the patches into healthy and Gleason grades 1 through 5 using 10×WSI,outperforming the state-of-the-art accuracy by 27%.Finally,the patient’s grade was determined by combining the patch-wise results.The method was also demonstrated for 4−class grading and binary classification of prostate cancer,achieving 93.0%and 99.6%accuracy,respectively.The reproducibility was over 90%.Since the proposedmethod determined the grades with higher accuracy and reproducibility using low-resolution images,it is more reliable and effective than existing methods and can potentially improve subsequent therapy decisions.
文摘目的旨在评估动态对比增强磁共振成像(dynamic contrast-enhancement magnetic resonance imaging,DCE-MRI)结合扩散加权成像(diffusion weighted imaging,DWI)在预测前列腺癌(prostate cancer,PCa)Ki-67表达和Gleason评分中的诊断效能。材料与方法回顾性分析了2019年1月至2023年10月自贡市第四人民医院收治的66例PCa患者的临床及影像资料。结合T2WI、DWI序列和由DWI自动计算出的表观扩散系数(apparent diffusion coeffieient,ADC),在DCE-MRI图像上手动勾画肿瘤感兴趣区(region of interest,ROI),计算ROI药代动力学参数,包括容积转运常数(volume transfer contrast,K^(trans))、速率常数(rate contrast,K_(ep))、血管外细胞外容积分数(extravascular extracellular volume fraction,Ve),并测量ADC值。根据靶向穿刺病理诊断Gleason评分和Ki-67表达水平,分为Ki-67高表达组(Ki-67>10%)和低表达组(Ki-67≤10%),Gleason评分低级别(GG 1~2)和高级别(GG 3~5)组。组间差异比较使用两独立样本t检验或非参数检验,采用Spearman相关分析评价DCE-MRI参数和ADC值与Ki-67、Gleason评分的相关性,并建立logistic回归模型,通过受试者工作特征(receiver operating characteristic,ROC)曲线评估诊断效能。结果ADC值与Ki-67表达、Gleason评分均呈负相关(P<0.001),K^(trans)、K_(ep)、Ve与Ki-67表达均呈正相关(P<0.001),K^(trans)、K_(ep)与Gleason评分均呈正相关(P<0.001)。Ki-67高、低表达组K^(trans)、K_(ep)、Ve、ADC值比较差异均具有统计学意义(P<0.01),Gleason评分高、低级别组K^(trans)、K_(ep)、ADC值比较差异均具有统计学意义(P<0.01);Ki-67表达的ROC曲线分析显示,联合模型K^(trans)+K_(ep)+Ve+ADC诊断效能最好,曲线下面积(area under the curve,AUC)为0.940;Gleason评分分级的ROC曲线分析显示,联合模型K^(trans)+K_(ep)+ADC诊断效能最好,AUC为0.861。结论DCE-MRI的药代动力学参数和ADC值相结合,在预测PCa的Ki-67表达和Gleason评分中显示出高诊断效能。联合使用DCE-MRI定量参数与ADC值可提高PCa病理分级和生物侵袭性的预测准确性。
