The use of artificial intelligence (AI) in diagnosing esophageal cancer shows promise in improving the accuracy and speed of diagnosis, according to study findings in the Diseases of the Esophagus.
The early endoscopic detection of macroscopic lesions is crucial in diagnosing esophageal cancer early. Because this diagnosis is largely subject to expert skill, providers have started to use AI to address potential human error during diagnosis. Researchers conducted a systematic review and meta-analysis to analyze the efficacy of endoscopic diagnosis of esophageal cancer.
Data were sourced from Pubmed, MEDLINE, and Ovid EMBASE databases. The researchers searched for articles and studies that were written in English and included endoscopic imaging. The use of AI to examine histology and AI-assisted endoscopic sponge cytology were reasons for exclusion.
A total of 48 articles were included in the analysis, and the analysis was divided into the AI diagnosis of esophageal squamous cell carcinoma and esophageal adenocarcinoma.
There were 24 studies that assessed the use of AI in the diagnosis of esophageal squamous cell carcinoma via endoscopy. Of these studies, 14 were included in the meta-analysis with 1590 patients used to validate the study cohort. Among the 14 studies, 6 used still images, 3 used video images, and 5 used a combination of both still and video images. The combined sensitivity of the AI diagnosis of esophageal squamous cell carcinoma was 91.2% (95% CI, 84.3%-95.2%) and specificity was 80% (95% CI, 64.3%-89.9%).
Of 12 studies that compared diagnostic ability between a range of skilled endoscopists and AI, 8 showed AI diagnosis was superior.
There were 5 studies that assessed the efficacy of AI assistance on endoscopist accuracy, sensitivity, and specificity. Among these studies, 1 study showed that using AI assistance improved the accuracy of novice endoscopists to a level comparable to expert endoscopists. Another study revealed that AI technology can improve expert endoscopist accuracy rates from 88.8% to 93.5%.
Of the 15 studies endoscopically diagnosing esophageal adenocarcinoma, the researchers reported the pooled sensitivity and specificity as 93.1% (95% CI, 86.8%-96.4%) and 86.9% (81.7%-90.7%), respectively.
“The value of accurate detection of early-stage malignancy offers patients the potential for curative treatment endoscopically, with survival outcomes similar to surgery, and a substantially improved Health Related Quality of Life through an organ preservation strategy,” study authors wrote. Study limitations include the detection of false negative and positive results, the use of still images not emulating real-life endoscopy, selection bias, and the lack of generalizability.
Guidozzi N, Menon N, Chidambaram S, Markar SR. The role of artificial intelligence in the endoscopic diagnosis of esophageal cancer: a systematic review and meta-analysis. Dis Esophagus. Published online July 21, 2023. doi:10.1093/dote/doad048