LUTs GET COLOURING: Optimization of Medical Image Perception in Multiple Modalities Using Colour Mapping and AI

Diagnostic errors in medical image analysis are responsible for 1 in 10 deaths in an outpatient setting and $17-29 billion in wasteful medical spending annually. As the majority of these errors are typically perceptual, developing a mechanism to increase the differentiation of background noise and signal from Regions of Interest (ROI) within medical images would serve to improve patient outcomes. Colour maps were generated using Lookup Tables (LUTs), which were either manipulated within a HSV (Hue, Saturation, Brightness Value) space to resemble parabolic functions, adjusted from preexisting LUTs from Fiji/ImageJ, or manually segmented to highlight anatomical structures. The efficacy of our generated LUTs was assessed with a survey (n=65) taken by non-radiologists and a Fiji-based automatic lesion detector we coded. While a greater sample size of qualified individuals and generated LUTs would be required to refine our results, five colour maps were shortlisted as being optimal for the detection of brain lesions on T1-weighted MRI scans.