Photo Credit: Libre de Droit

A meta-analysis on machine learning in early gastric cancer diagnosis found that ML models showed high sensitivity (0.91) and specificity (0.85), improving diagnostic accuracy for non-specialist clinicians, highlighting their potential in broad clinical applicability during endoscopy.

The following is a summary of “Value of Machine Learning Approaches in the Diagnosis of Early Gastric Cancer: A Systematic Review and meta-analysis,” published in the February 2024 issue of Oncology by Shi et al.

The increasing interest in applying machine learning (ML) to identify early gastric cancer (EGC) prompted this systematic review and meta-analysis to evaluate the specific diagnostic performance of image-based ML in EGC diagnosis. A comprehensive search of PubMed, Embase, Cochrane Library, and Web of Science up to September 25, 2022, was conducted, with the risk of bias assessed using QUADAS-2. 

The meta-analysis, employing a bivariant mixed-effect model, revealed that ML-based models demonstrated a sensitivity (SEN) of 0.91 (95% CI: 0.87–0.94), specificity (SPE) of 0.85 (95% CI: 0.81–0.89), and summary receiver operating characteristic (SROC) of 0.94 (95% CI: 0.39–1.00) in the training set and 0.90 (95% CI: 0.86–0.93), 0.90 (95% CI: 0.86–0.92), and 0.96 (95% CI: 0.19–1.00) in the validation set. For non-specialist clinicians, EGC diagnosis had a SEN of 0.64 (95% CI: 0.56–0.71), SPE of 0.84 (95% CI: 0.77–0.89), and SROC of 0.80 (95% CI: 0.29–0.97), whereas specialist clinicians showed a SEN of 0.80 (95% CI: 0.74–0.85), SPE of 0.88 (95% CI: 0.85–0.91), and SROC of 0.91 (95% CI: 0.37–0.99). With the ML model’s assistance, the SEN of non-specialist physicians significantly improved from 0.64 to 0.76 in EGC diagnosis. 

In conclusion, ML-based diagnostic models exhibit superior performance in identifying EGC, and they can enhance the diagnostic accuracy of non-specialist clinicians to the level of specialists. These findings underscore the potential of ML models to assist less experienced clinicians in diagnosing EGC during endoscopy, indicating broad clinical applicability.

Source: wjso.biomedcentral.com/articles/10.1186/s12957-024-03321-9