Accuracy of diminutive colorectal polyp diagnosis when using pathology alone, computer aided characterisation alone or a combined approach

Aims Multiple steps between polyp detection and obtaining a final pathology report can result in an incorrect diagnosis. These steps include resection and retrieval, during which specimens may fracture or be lost, followed by processing (embedding and sectioning) in the pathology laboratory, and final analysis, all of which can lead to errors and misdiagnoses. These steps have previously not been taken into account when estimating the true accuracy of pathology within an “intention-to-diagnose” framework. Intra-colonoscopy computer-aided characterization (CADx) has emerged as an alternative strategy and has the benefit of not requiring specimen retrieval and handling. We were interested in pragmatically comparing Pathology-based diagnosis to CADx from an “intention-to-diagnose” framework.

Methods We conducted a post-hoc analysis of a prospective clinical study. Primary outcome was accuracy in polyp diagnosis when using CADx or pathology alone, taking into account the impact of non-resection, non-retrieval, and misdiagnosis when using pathology. Secondary outcome was the accuracy of a combined strategy where CADx is used for arbitration of cases that could not be diagnosed in pathology. For the pathology strategy, unresected polyps, unretrieved polyps, polyps not received in pathology, and diagnoses of “mucosal fold” were considered incorrect. In cases of “mucosal fold” diagnosis, 3 expert endoscopists (HP, DKR, CH) evaluated videos of these lesions to confirm whether a polyp was truly present. For the CADx strategy, accuracy was compared to pathology as gold standard when pathology was available and extrapolated to the polyps with no pathology diagnosis.

Results A total of 467 diminutive polyps in 269 consecutive patients were included. Pathology, when taking into account cases where diagnosis could not be obtained (not resected, not retrieved, not received by pathology lab) and cases with inaccurate diagnosis (mucosal fold with 3 expert endoscopist video confirmation of the presence of polyp), had an estimated accuracy of 77.1% (95%CI 73.0-80.8). We found that CADx resulted in an accurate diagnosis in 74.7% (95%CI 69.7-79.3) when used alone. Inaccurate diagnoses in the pathology strategy were: 1.7% not resected due to losing sight of the polyp during endoscopy or other factors, resulting in no available histologic information; 6.4% retrieved despite attempts at retrieval; 1.1% were retrieved but not received by the pathology unit; and 13.7% were diagnosed as normal mucosa/mucosal folds. When using a combined approach of pathology with CADx for arbitration, the accuracy increased to 94.2%.

Conclusions Our findings indicate that using pathology alone for diagnosis of colorectal polyps leads to unavailable or inaccurate diagnosis in 22.9% of cases with an overall accuracy of only 77%. Comparatively, CADx alone showed similar accuracy as diagnosis can be performed in all detected polyps. A combined approach allowed for the highest diagnostic accuracy.

Publication History

Article published online:
27 March 2025