Bridging the Disciplinary gap for Computer Aided Diagnosis in Colonoscopy: Detection, Evaluation and Diagnostics

Every year, more than 4000 people get diagnosed with colorectal cancer in Norway. Early disease detection and treatment has a huge impact on 5-year survival, going from a low 10-30% if detected in later stages to a high 90% in early stages. The gold standard for diagnosing colon cancer is through a camera examination of the large bowel called colonoscopy, looking for polyps which may evolve or have already evolved into cancer. There are large differences in polyp detection rates between doctors, and low detection rates are associated with increased risk of cancer. This project's goal is to develop a Computer-aided Diagnosis (CADx) system using Artificial Intelligence (AI). The system can recognise and classify both polyps and other lesions in real-time during colonoscopies. Our goal is to largely eliminate the rate of missed lesions in such examinations and thus increase the number of detected lesions, possibly preventing cancer. As a method for creating the AI models, we will use deep neural networks, algorithms shown to be effective in image and video analyses. The system will cover the whole pipeline, from extracting videos from the colonoscopy device, through real-time video analyses, to immediate feedback to the clinicians. Immediate feedback during the examination is an important part of the project, to allow the doctors to closer assess the detected findings without delay. To make sure the results of the analyses are relatable, we will look into different explainable methods for evaluating and presenting the conclusions, making sure the results are trustworthy. Increased polyp detection rate will reduce the risk of colorectal cancer. We believe this can be an important asset to colonoscopy, and benefit many patients.

-

The main objective of this Ph.D. project is to create and evaluate a Computer-aided Diagnosis (CADx) system using Artificial Intelligence (AI) to recognise and classify polyps and other pathological findings in real-time during colonoscopies, and researching understandable methods for evaluating the results. As a method for creating the AI models, we will use deep neural networks (DNNs) which are state of the art in image and video related analysis. The goal is to largely eliminate the rate of missed lesions in such examinations and thus increase the number of detected lesions, possibly preventing it from evolving into cancer. Our system is covering the whole pipeline, from extracting videos from the colonoscopy rack, through real-time detection and classification of lesions, to immediate feedback to the clinician. The intermediate feedback during the assessment of the patient is an important part of the project since it will allow the clinician to react immediately on the detected findings. To make sure the results of the analysis is relatable for all the involved disciplines, we will research and develop different ways to evaluate the results, this includes quantifying the effect in clinical trials. We will also look into automatic report generation, with suggested text and images for the medical records, saving time for the doctor.