Abstract

The aim of this challenge is to explore methods that enhance the explainability of Odin-Vision’s current machine learning models to aid clinical decision-making. Their current capabilities include a real-time detection and classification model, deployed in a clinical setting, where a polyp is first imaged by a clinician and automatically classified as adenoma or non-adenoma; a binary classification task. The procedure is time sensitive and each polyp gets imaged approximately only once, with clinicians taking a few seconds for image capture and decision making. The aim of the machine learning model is to aid the clinician’s decision process, providing confidence in more ambiguous cases and substantially increase the reproduciblity of those decisions. The clinician’s trust in the model is also particularly important to encourage widespread uptake and acceptance of automated methods in a clinical setting.

 

Citation information

Data Study Group team. (2021, November 26). Data Study Group Final Report: Odin Vision. Zenodo. https://doi.org/10.5281/zenodo.5562660

Additional information

PI – Paul Duckworth