University College Dublin Researchers Harness Computer Vision and AI for Real-Time Biopsy-Free Cancer Discrimination
Innovative Approach Enables Quicker Patient Diagnosis
“MATLAB tools provide a means to develop and utilize advanced machine learning methods without having to possess beyond basic programming abilities.”
Key Outcomes
- MATLAB enabled medical professionals to jump-start their AI research with minimal coding experience
- Classification Learner app helped achieve accuracy levels of 90% when assessing 50 lesions for cancerous tissue
- Full field of view allowed for increased sampling and created cleaner data sets by identifying noise—from surgical instruments, for example—for removal
Using AI tools from MATLAB, researchers were able to assess colon polyps without taking a biopsy sample.
The University College Dublin (UCD) Centre for Precision Surgery is a research institute in a medical school that focuses on using technology to advance patient health. One major challenge that UCD researchers have addressed is how to effectively assess a patient’s potential for colorectal cancer in real time. Colorectal cancer is the fourth leading cause of cancer-related death worldwide and traditionally relies on a diagnosis via the biopsy of a colorectal polyp. However, assessing a biopsy sample is time-consuming and provides only a partial look at the polyp. Instead, UCD researchers used MATLAB® tools to enable AI-powered results for polyp analysis in real time, with a full field of view to assess the entire polyp.
Today, surgical teams employ medical dyes amplified by surgical video cameras to assess tissues near cancerous regions. Using App Designer and Computer Vision Toolbox™, the team developed an app to stabilize these videos, so they could compute features from the extracted fluorescence intensity time histories representing the dye perfusion. Labels were then applied to video images to mark cancerous and benign areas, as predetermined by clinician expertise and biopsy results.
Additionally, using Classification Learner app, the labeled features were then used to train a machine learning model to discriminate cancerous regions for new surgical videos without human assistance. The trained model outperformed clinicians in identifying cancerous tissue and performed at least as well as a traditional biopsy, the current pre-excision gold standard.
Easy-to-use MATLAB tools allowed UCD researchers to jump-start their AI research without prior coding knowledge. This approach could lead to quicker patient diagnosis and better clinical outcomes in the future, and the research team plans to use MATLAB tools to assess plastic surgery grafts and bowel resection as well.
Related Resources
