Nicola Rosberg (CRT AI), Simone Innocente (Tyndall), Barry O’Sullivan (Insight UCC), Andrea Visentin (Insight UCC)
Automating the diagnosis of various frequently occurring afflictions presents one approach to supporting healthcare professionals. However, several problems, including high inter-variability between patients, low explainability of decision-making algorithms and professional reluctance to rely on new technology, are interfering with the implementation of these algorithms.
To tackle these barriers, our group at the Insight Research Ireland Centre for Data Analytics at UCC and Tyndall National Institute is exploring the feasibility of using spectroscopy in combination with explainable machine learning models to allow the automatic identification of a series of conditions, including cancer. Through the use of light signals, distinctive tissue signatures can be identified, which can, in turn, be classified by a trained machine learning algorithm. Our research shows that the speed and accuracy of the developed algorithm may allow it to successfully be integrated into a hospital setting to provide supporting diagnoses and inform medical decision-making. However, medically suited, reliable explanations are needed to qualify for hospital integration, which is a topic that remains challenging. In addition, large, diverse datasets are needed for training unbiased models, which are challenging to collect in the medical domain.
If successful, the developed algorithms have the potential to ease medical workflows, decrease hospital load and support practitioners in making high-importance diagnoses, ultimately increasing the efficiency of the medical system and improving patient health.