Reliable classification of polyps based on artificial intelligence: a development and validation study - PubMed
4 hours ago
- #Colorectal Cancer
- #Digital Pathology
- #Artificial Intelligence
- POLARIS is an AI-based tool developed to assist pathologists in classifying colorectal polyps, addressing the shortage of pathologists.
- The tool was trained using 15,079 whole-slide images from 2993 patients in the UK bowel cancer screening program (2014-2018).
- POLARIS categorizes polyps into five classes based on malignancy risk, regrouping them into two broader categories for clinical interpretation.
- External validation on 10,842 WSIs showed POLARIS correctly identified 98.94% of high-risk polyps (HGD and adenocarcinoma) and 83.04% of low-risk polyps (normal tissue and tubular adenomas with LGD).
- The model achieved a balanced accuracy of 86.65% in external validation and an AUROC of 0.9449 for distinguishing low-risk from high-risk polyps.
- Pathologists agreed with POLARIS over clinical diagnosis in 92.5% of reviewed cases where predictions differed.
- POLARIS has potential to improve diagnostic workflows by reducing the number of slides requiring pathologist review and highlighting high-risk regions.
- The development was funded by The Norwegian Cancer Society, with several authors having potential conflicts of interest related to commercialization.