Artificial Intelligence Shows Promise for Detecting Early Oesophageal Squamous Cell Carcinomas
By Denise Baez
Artificial intelligence (AI) can help endoscopists diagnose precancerous lesions and early oesophageal squamous cell carcinomas (ESCCs), according to a study published in Gastrointestinal Endoscopy.
“We developed a system for computer-assisted diagnosis (CAD) for real-time automated diagnosis of precancerous lesions and early ESCCs to assist the diagnosis of oesophageal cancer,” wrote LinJie Guo, MD, West China Hospital, Sichuan University, Chengdu, China, and colleagues.
The researchers used 6,473 narrow-band imaging (NBI) images, including precancerous lesions, early ESCCs, and noncancerous lesions, to train the CAD system. It was validated using both endoscopic images and video datasets. An artificial intelligence probability heat map was generated for each input of endoscopic images, with yellow indicating a high possibility of a malignant lesion and blue indicating noncancerous lesions. When the CAD system detected any precancerous lesions or early ESCCs, the lesion of interest was masked with colour.
The image datasets contained 1,480 malignant NBI images from 59 consecutive cancerous cases, which yielded a sensitivity of 98.04%, and 5,191 noncancerous NBI images from 2,004 cases with a specificity of 95.03%. The area under the curve was 0.989.
For video datasets of non-magnifying videos, per-frame sensitivity was 60.8% and per-lesion sensitivity (defined as the percentage of lesions correctly detected by the algorithm in at least 1 frame) was 100%, whereas for magnifying videos. The per-frame sensitivity was 96.1% and per-lesion sensitivity was 100%. Unaltered, full-range normal oesophagus videos included 33 videos (per-frame specificity, 99.9%; per-case specificity, 90.9%). The model was capable of processing at least 25 frames per second, with a latency period of
In an accompanying editorial, Shyam J. Thakkar, MD, and Gursimran S. Kochhar, MD, Center for Advanced Endoscopy, Pittsburgh, Pennsylvania, wrote: “Artificial intelligence with the use of deep learning provides an opportunity to improve the detection rate of early oesophageal cancer lesions, particularly by junior endoscopists. The model described by Guo et al had high diagnostic sensitivity and specificity, not only in image analysis but also in real-time video settings. This is an important study that paves the way for the development of better models for detection of oesophageal cancer lesions in real-time clinical settings, but it will require randomised prospective clinical trials for validation.”
SOURCE: Gastrointestinal Endoscopy