ABSTRACT
Accurate segmentation of whole slide images is of great significance for the diagnosis of pancreatic cancer. However, developing an automatic model is challenging due to the complex content, limited samples, and high sample heterogeneity of pathological images. This paper presented a multi-tissue segmentation model for whole slide images of pancreatic cancer. We introduced an attention mechanism in building blocks, and designed a multi-task learning framework as well as proper auxiliary tasks to enhance model performance. The model was trained and tested with the pancreatic cancer pathological image dataset from Shanghai Changhai Hospital. And the data of TCGA, as an external independent validation cohort, was used for external validation. The F1 scores of the model exceeded 0.97 and 0.92 in the internal dataset and external dataset, respectively. Moreover, the generalization performance was also better than the baseline method significantly. These results demonstrate that the proposed model can accurately segment eight kinds of tissue regions in whole slide images of pancreatic cancer, which can provide reliable basis for clinical diagnosis.
Subject(s)
Humans , China , Pancreatic Neoplasms/diagnostic imaging , LearningABSTRACT
Objective:The tumor-stroma ratio(TSR)is considered an independent prognostic factor for gastric cancer.Traditionally,TSR as-sessments have relied on the visual evaluation of surgical specimens,which is a method that lacks objectivity.This study was conducted to investigate whether the TSR in preoperative biopsy specimens can be automatically quantified using deep learning methods and whether the TSR value can be used to predict the efficacy of neoadjuvant chemotherapy(NAC)in patients with gastric cancer.Methods:In total,148 preoperative biopsy slides and 43 surgical resection slides from patients with gastric cancer who underwent NAC treatment at Yunnan Can-cer Hospital between March 2013 and March 2020 were used in the study.Tumor region segmentation and epithelial-stromal segmentation models were developed.The surgical resection slides were used to trained and evaluate the model,and the biopsy slides were used to test their predictive abilities.The TSR values were determined on the basis of the intersection of predictions from both models.The postoperat-ive pathological tumor regression grade(TRG)was used to categorize patients into good responders(TRG 0-1)and poor responders(TRG 2-3).Univariate and multivariate Logistic regression analyses were conducted to determine the correlation between the TSR value and the ef-ficacy of NAC in gastric cancer.Results:The intersection over union(IOU)value was 0.94 for the tumor tissue segmentation model and 0.88 for the epithelial-stromal segmentation model.Using cutoff values of 44.93％and 70.22％,patients were classified into low,intermediate,and high TSR groups.The proportion of good responders was significantly different among these groups(P＜0.05).Multivariate Logistic re-gression analysis indicated that the TSR was an independent predictor of NAC response in gastric cancer(OR=0.10,95％CI:0.03-0.32).When the TSR three-category classification was added as a predictor of treatment response alongside conventional clinical information,the area under curve(AUC)increased from 0.71 to 0.85.Conclusions:This deep learning model is capable of automatically segmenting tumor,epi-thelial,and stromal regions based on pathological slides,accurately calculating TSR value,and predicting the efficacy of NAC on the basis of the automatically computed TSR values.