Autonomous CADx Program Outperformed Experienced Endoscopists in Assigning Postpolypectomy Surveillance Intervals
Douglas K. Rex, MD, MASGE, reviewing Djinbachian R, et al. Gastroenterology 2024 Feb 6.
In the study, 467 patients were randomized to receive computer-aided diagnosis (CADx) predictions of diminutive colorectal polyp pathology by an artificial intelligence (AI) program alone, without any assistance or input from a human or other endoscopist, versus prediction by the endoscopist with input from the AI program.
The optical diagnosis accuracy for autonomous AI was numerically higher than in the arm with endoscopist input (77.2% vs 72.1%). Predicted surveillance intervals compared to pathology-based intervals had a higher agreement for autonomous AI than with the arm with endoscopist input (91.5% vs 82.1%; P = .016). Intervals selected in the arm with the endoscopist input were more likely to be too short (16.2% vs 7.1%), which mostly appears to have resulted from the endoscopist predicting some lesions to be sessile serrated lesions (SSLs) and the autonomous AI program never choosing SSL as the type of lesion.
COMMENT
The CADx accuracy in this study was not better than CADx in previously published studies. What is novel here is testing the autonomous performance of a CADx without human input and the result of a real-time CADx program outperforming experienced endoscopists with regard to any major outcome. However, the latter appears to be an artifact of allowing the endoscopists to predict SSL in diminutive polyps. There is currently no accurate endoscopic classification system for differentiating SSLs from hyperplastic polyps in diminutive polyps and no CADx program capable of this distinction.
Note to readers: At the time we reviewed this paper, its publisher noted that it was not in final form and that subsequent changes might be made.
Douglas K. Rex, MD, MASGE
CITATION
Djinbachian R, Haumesser C, Taghiakbari M, et al. Autonomous artificial intelligence versus AI assisted human optical diagnosis of colorectal polyps: a randomized controlled trial. Gastroenterology 2024 Feb 6. (Epub ahead of print) (https://doi.org/10.1053/j.gastro.2024.01.044)
COMMENT
This study affirms important principles for FIT screening. First, cancers detected by FIT remain in an early stage during subsequent screening rounds. Further, increasing the cutoff level of hemoglobin for positive test results reduces sensitivity but does not shift the fraction of detected cancers substantially toward later-stage CRC. Thus, programs can adjust the cutoff to levels appropriate for their colonoscopy resources, and the detected cancers will be in an early stage across a range of hemoglobin thresholds for positive tests.