Categories
Change Password!
Reset Password!
Computer-aided diagnosis assistance does not improve the sensitivity of optical diagnosis, and while specificity slightly increases, it remains suboptimal.
A new study issued in "Annals of Internal Medicine" has examined the use of computer-aided diagnosis (CADx) in predicting the histology of small colorectal polyps in real-time, aiming to minimize unnecessary resections and pathologic evaluations, thus lowering colonoscopy costs. Previous research indicated that while CADx was highly accurate, it did not outperform expert endoscopists.
The prospective, multicenter study, conducted across 6 centers, involved 1252 patients and 49 general endoscopists. The goal was to assess the diagnostic performance of histologic predictions before and after CADx assistance in a real-world setting. The key endpoints included the sensitivity and specificity of histologic predictions for adenomas measuring 5 mm or less, both with and without CADx assistance. Additional estimates were also offered based on the polyp location and confidence level for clinical use.
Results revealed that CADx offered a diagnosis for 96% of polyps measuring 5 mm or less, with no vital difference in sensitivity between unassisted and CADx-assisted groups (90.7% vs. 90.8%). However, specificity was notably higher in the CADx-assisted group (64.7% vs. 59.5%). Additionally, a large proportion of polyps could be safely resected and discarded without further pathologic evaluation, with 88.2% in the CADx-assisted group and 86.1% in the unassisted group.
Among rectosigmoid polyps, nearly half (49.5%) in the CADx-assisted group could remain in situ without resection, compared to 47.9% in the unassisted group. While the study highlighted promising benefits in minimizing unnecessary procedures, CADx did not markedly improve the sensitivity of optical diagnosis. Despite a slight increase in specificity, the CADx-assisted diagnosis remained suboptimal. The study cautions that clinical decision-making based on CADx might vary outside of a controlled trial setting.
Annals of Internal Medicine
Artificial Intelligence for Real-Time Prediction of the Histology of Colorectal Polyps by General Endoscopists
Douglas K Rex et al.
Comments (0)