Towards early bowel cancer detection: A data driven dynamic method of lesions characterisation using a robotic capsule

This is the authors' abstract. We don't add key points for this paper.

Bowel cancer (BC) ranks as the second deadliest cancer globally with about a million deaths annually [1]. The high mortality is partly due to late stage diagnosis as current methods rely on visual examinations that are not reliable for detecting hard-to-visualise early-stage BCs. A notable difference between early-stage malignant lesions and their benign counterparts is their contrasting biomechanical properties, such as stiffness. In this study, a new approach to BC diagnosis is proposed. The approach uses a two-stage machine learning (ML) proce- dure to characterise bowel lesions based on the resulting dynamics of a recently developed self-propelled vibro- impact capsule (SP-VIC) travelling and encountering lesions in the bowel. As an impacting system, the resulting dynamics of the SP-VIC are intensely nonlinear and quite sensitive to any form of perturbation in its system parameters and its contacted tissue properties, such as stiffness. This sensitivity is novelly explored for bowel lesions characterisation using ML as a data driven approach in the absence of a mathematical hypothesis.