Premium
Deep learning analysis for the detection of pancreatic cancer on endosonographic images: a pilot study
Author(s) -
Tonozuka Ryosuke,
Itoi Takao,
Nagata Naoyoshi,
Kojima Hiroyuki,
Sofuni Atsushi,
Tsuchiya Takayoshi,
Ishii Kentaro,
Tanaka Reina,
Nagakawa Yuichi,
Mukai Shuntaro
Publication year - 2021
Publication title -
journal of hepato‐biliary‐pancreatic sciences
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.63
H-Index - 60
eISSN - 1868-6982
pISSN - 1868-6974
DOI - 10.1002/jhbp.825
Subject(s) - overdiagnosis , medicine , pancreatic cancer , pancreatic ductal adenocarcinoma , receiver operating characteristic , cad , pancreas , univariate analysis , multivariate analysis , pancreatitis , cancer , artificial intelligence , radiology , pathology , computer science , biology , biochemistry
Background/Purpose The application of artificial intelligence to clinical diagnostics using deep learning has been developed in recent years. In this study, we developed an original computer‐assisted diagnosis (CAD) system using deep learning analysis of EUS images (EUS‐CAD), and assessed its ability to detect pancreatic ductal carcinoma (PDAC), using control images from patients with chronic pancreatitis (CP) and those with a normal pancreas (NP). Methods A total of 920 endosonographic images were used for the training and 10‐fold cross‐validation, and another 470 images were independently tested. The detection abilities in both the validation and test setting were assessed, and independent factors associated with misdetection were identified among participants' characteristics and endosonographic image features. Results Regarding the detection ability of EUS‐CAD, the areas under the receiver operating characteristic curve were found to be 0.924 and 0.940 in the validation and test setting, respectively. In the analysis of misdetection, no factors were identified on univariate analysis in PDAC cases. On multivariate analysis of non‐PDAC cases, only mass formation was associated with overdiagnosis of tumors. Conclusions Our pilot study demonstrated the efficacy of EUS‐CAD for the detection of PDAC.