Development of artificial intelligence for the detection and staging of esophageal cancer

Abstract

: Computer-assisted diagnosis (CAD) using deep learning, based on convolutional neural networks (CNNs), is rapidly developing in modern society and clinical practice. In the medical field of endoscopy, a CAD system can be used in the detection and staging of superficial esophageal cancer. In the detection of superficial esophageal squamous cell carcinoma (ESCC), several studies were reported as using still images of conventional white-light endoscopy and narrow-band imaging (NBI). The sensitivity of CAD systems is over 90%, and in some reports, higher than that of expert endoscopists. In addition, there is a report using video for a validation set that showed good performance. In diagnosing invasion depth, there are reports using conventional white-light imaging, NBI, and magnifying endoscopy with NBI using still images; there are also reports of intrapapillary capillary loop pattern identification using still images. Using white-light endoscopy and NBI, CAD systems showed sensitivity of 84.1–95.4%, specificity of 73.3–79.2%, and accuracy of 80.9–92.9% for differentiating SM1 cancers from SM2 or SM3 cancers in pathology. Additional systems could accurately classify intrapapillary capillary loop patterns as normal or abnormal, with comparable performance to experienced endoscopists. As for esophageal adenocarcinoma (EAC) that arises in Barrett’s esophagus (BE), there are detection reports that show favorable performance. Furthermore, the CAD system from one report identified the location suitable for biopsy; this is helpful for endoscopists because of the difficulty in determining the location of early EAC in BE. Although these studies are still only at research level, excellent performance has been achieved for detecting and staging of superficial esophageal carcinoma by CAD systems. In the near future, CAD systems will support us in detecting and staging esophageal cancers in daily clinical practice, leading to a better prognosis.