Introduction
Artificial intelligence (AI) has recently made remarkable progress, and computer-aided detection (CAD) system for gastrointestinal endoscopy is rapidly evolving. The development of AI software that provides real-time information displayed directly on an endoscopic screen is expected to aid in the detection of neoplastic lesions. Furthermore, the integration of AI with virtual chromoendoscopy such as blue laser/light imaging (BLI) or linked color imaging (LCI) is expected to improve the accuracy of detecting early stage cancer lesions. Based on this background, Fujifilm recently developed a CAD system (CAD EYE, Fujifilm, Tokyo, Japan) for the detection of gastrointestinal neoplasms using white-light imaging (WLI) and image-enhanced endoscopy (IEE).
Aims & Methods
This study aimed to validate the performance of CAD in identifying esophageal squamous cell carcinoma (ESCC) and gastric neoplasm (GN). This multicenter retrospective study utilized image datasets comprising 15 consecutive video frames captured using WLI, BLI, and LCI. The sensitivity and specificity of CAD for successful detection were calculated using the gold standard, which consists of image datasets of neoplastic lesions annotated by experienced endoscopists.
Results
A total of 620, 679, and 682 ESCC datasets were analyzed in the WLI, BLI, and LCI groups, respectively. The sensitivity and specificity of ESCC detection were 85.9% (176/205) and 93.3% (387/415) in the WLI group, 97.6% (206/211) and 92.9% (435/468) in the BLI group, and 96.6% (201/208) and 93.2% (442/474) in the LCI group. The detection rates for pT1a ESCC were 85.3% (145/170), 97.3% (181/186), and 97.2% (172/177) in the WLI, BLI, and LCI, respectively. For GN, 846 WLI and 885 LCI datasets were analyzed. The sensitivity and specificity of GN detection were 95.5% (231/242) and 85.4% (516/604) in the WLI group, and 93.9% (263/280) and 93.9% (568/605) in the LCI group, respectively. The detection rates for pT1a early gastric cancer were 93.8% (152/162) and 92.4% (182/197) in the WLI and LCI groups, respectively.
Conclusion
The CAD system demonstrated high sensitivity in detecting ESCC and GN, highlighting its potential as a promising tool for clinical applications.
Disclosure
Seiichiro Abe: Endoscopy Division, National Cancer Center Hospital, Tokyo, Japan
Yoshiyasu Kitagawa: Endoscopy Division, Chiba Cancer Center, Chiba, Japan
Waku Hatta: Division of Gastroenterology, Tohoku University Graduate School of Medicine, Miyagi, Japan
Takao Maekita: Second Department of Internal Medicine Wakayama Medical University, Wakayama, Japan
Motohiko Kato: Center for Diagnostic and Therapeutic Endoscopy, Keio University School of Medicine, Tokyo, Japan
Akihito Nagahara and Hiroya Ueyama: Department of Gastroenterology, Juntendo University Faculty of Medicine, Tokyo, Japan
Hiroyuki Osawa: Department of Medicine, Division of Gastroenterology, Jichi Medical University, Shimotsuke, Japan
Osamu Dohi: Molecular Gastroenterology and Hepatology, Graduate School of Medical Science, Kyoto Prefectural University of Medicine, Kyoto, Japan
Hirotaka Nakashima: Department of Gastroenterology, Foundation for detection of early gastric carcinoma, Tokyo, Japan
Cancer screening center, Cancer institute hospital of Japanese foundation for cancer research, Tokyo, Japan
Kazuhiro Furukawa: Department of Gastroenterology and Hepatology, Nagoya University Graduate School of Medicine, Aichi, Japan
Shiro Oka: Department of Gastroenterology, Hiroshima University Hospital, Hiroshima, Japan
Tomoko Yokoyama: FUJIFILM Nishiazabu endoscopy clinic, Tokyo, Japan
Toru Ito: Department of Gastroenterological Endoscopy, Kanazawa Medical University Hospital, Ishikawa, Japan
Ichiro Oda: Endoscopy Division, National Cancer Center Hospital, Tokyo, Japan
Department of Internal Medicine, Kawasaki Rinko General Hospital, Kanagawa, Japan