A deep learning model to detect pancreatic ductal adenocarcinoma on endoscopic ultrasound-guided fine-needle biopsy

Abstract

Histopathological diagnosis of pancreatic ductal adenocarcinoma (PDAC) on endoscopic ultrasonography-guided fine-needle biopsy (EUS-FNB) specimens has become the mainstay of preoperative pathological diagnosis. However, on EUS-FNB specimens, accurate histopathological evaluation is difficult due to low specimen volume with isolated cancer cells and high contamination of blood, inflammatory and digestive tract cells. In this study, we performed annotations for training sets by expert pancreatic pathologists and trained a deep learning model to assess PDAC on EUS-FNB of the pancreas in histopathological whole-slide images. We obtained a high receiver operator curve area under the curve of 0.984, accuracy of 0.9417, sensitivity of 0.9302 and specificity of 0.9706. Our model was able to accurately detect difficult cases of isolated and low volume cancer cells. If adopted as a supportive system in routine diagnosis of pancreatic EUS-FNB specimens, our model has the potential to aid pathologists diagnose difficult cases.

Weakly supervised and transfer learning for adenocarcinoma classification in transurethral resection of the prostate whole slide images

“…Prior to the training of prostate adenocarcinoma model using TUR-P WSIs, we have demonstrated the existing adenocarcinoma classification models AUC performances on TUR-P test sets (Table 2). Existing adenocarcinoma classification models were summarized in Table 3: (1) breast invasive ductal carcinoma (IDC) classification model (Breast IDC (x10, 512)) Kanavati and Tsuneki (2021a), (2) breast invasive ductal carcinoma and ductal carcinoma in-situ (DCIS) classification model (Breast IDC, DCIS (x10, 224)) Kanavati et al (2022), (3) colon adenocarcinoma (ADC) and adenoma (AD) classification model (Colon ADC, AD (x10, 512)) Iizuka et al (2020), (4) colon poorly differentiated adenocarcinoma classification model (transfer learning model from stomach poorly differentiated adenocarcinoma classification model) (Colon poorly ADC-1 (x20, 512)) Tsuneki and Kanavati (2021), (5) colon poorly differentiated adenocarcinoma classification model (EfficientNetB1 trained model) (Colon poorly ADC-2 (x20, 512)) Tsuneki and Kanavati (2021), (6) stomach adenocarcinoma and adenoma classification model (Stomach ADC, AD (x10, 512)) Iizuka et al (2020), (7) stomach poorly differentiated adenocarcinoma classification model (Stomach poorly ADC (x20, 224)) Kanavati and Tsuneki (2021b), (8) stomach signet ring cell carcinoma (SRCC) classification model (Stomach SRCC (x10, 224)) Kanavati et al (2021a), (9) pancreas endoscopic ultrasound guided fine needle aspiration (EUS-FNA) biopsy adenocarcinoma classification model (Pancreas EUS-FNA ADC (x10, 224)) Naito et al (2021), and (10) lung carcinoma classification model (Lung Carcinoma (x10, 512)) Kanavati et al (2020). Table 3 shows that Colon poorly ADC-2 (x20, 512) and Lung Carcinoma (x10, 512) models exhibited both high ROC-AUC and low log loss values as compared to other models.…”

PancreaSys: An Automated Cloud-Based Pancreatic Cancer Grading System

“…This is supported by the latest review article by the British Journal of Cancer on deep learning in cancer pathology (Echle et al, 2021) covering various cancer types but not pancreatic cancer. However, we found a very recent study on detection and classification of pancreatic adenocarcinoma in WSIs using DL with EfficientNet-B1 architecture (Naito et al, 2021). This model was pre-trained using ImageNet, and the analysis for transfer learning of 372 WSIs was done using overlapping fixed-sized tiles of 512 by 512 pixels with a stride of 256 pixels.…”

Breast Invasive Ductal Carcinoma Classification on Whole Slide Images with Weakly-Supervised and Transfer Learning

“…We refer to the trained models as TL <magnification> <tile size> <model size>, based on the different configurations. As we had at our disposal six models [18,27,[34][35][36][37] that had been trained specifically on specimens from different organs (stomach, colon, lung, and pancreas), we evaluated those models without fine-tuning on the test sets to investigate whether morphological cancer similarities transfer across organs without additional training.…”