A new study highlighted the outcomes of an artificial intelligence (AI) system in detecting upper gastrointestinal cancers.
“Upper gastrointestinal cancers (including oesophageal cancer and gastric cancer) are the most common cancers worldwide,” the researchers wrote. “Artificial intelligence platforms using deep learning algorithms have made remarkable progress in medical imaging but their application in upper gastrointestinal cancers has been limited.”
The study authors, whose work appeared in The Lancet Oncology, described their success using the Gastrointestinal Artificial Intelligence Diagnostic System (GRAIDS) to diagnose these cancers by analyzing clinical endoscopy imaging data.
The case-control, diagnostic study spanned six Chinese hospitals of different tiers (e.g. municipal, provincial, and national). Consecutive patients aged ≥ 18 years without a past endoscopy were recruited for the study. Patients were eligible for study participation if they had upper gastrointestinal cancer lesions (including esophageal and gastric cancers) that were histologically proven malignancies. Images from the Sun Yat-sen University Cancer Center were randomized 8:1:1 to the training and intrinsic verification datasets for developing GRAIDS, and the internal validation dataset for evaluating GRAIDS performance. An internal and prospective validation set was used to measure GRAIDS’ diagnostic performance, as were five more external validation sets from five primary care hospitals. GRAIDS was also compared to endoscopists with expert, competent, and trainee degrees of expertise.
A total of 1,036,496 endoscopy images (84,424 gastrointestinal cancer patients) were used in the development and testing of GRAIDS. GRAIDS’ diagnostic accuracy was 0.955 (95% confidence interval [CI] 0.952–0.957) in the internal validation set, 0.927 (0.925–0.929) in the prospective set, and was between 0.915 (0.913–0.917) and 0.977 (0.977–0.978) in the five external validation sets. When compared to the endoscopists, diagnostic sensitivity was similar between GRAIDS and the expert (0.942 [95% CI 0.924–0.957] vs. 0.945 [0.927–0.959]; P = 0.692), and GRAIDS was superior to the competent (0.858 [0.832–0.880], P < 0.0001) and trainee (0.722 [0.691–0.752], P < 0.0001) endoscopists. The positive predictive values for GRAIDS and the expert, competent, and trainee endoscopists were 0.814 (95% CI 0.788–0.838), 0.932 (0.913–0.948), 0.974 (0.960–0.984), and 0.824 (0.795–0.850), respectively. The negative predict value for GRAIDS was 0.978 (95% CI 0.971–0.984) compared to 0.980 (0.974–0.985) for the expert, 0.951 (0.942–0.959) for the competent, and 0.904 (0.893–0.916) for the trainee endoscopists.
The researchers concluded, “GRAIDS achieved high diagnostic accuracy in detecting upper gastrointestinal cancers, with sensitivity similar to that of expert endoscopists and was superior to that of non-expert endoscopists. This system could assist community-based hospitals in improving their effectiveness in upper gastrointestinal cancer diagnoses.”