Author: Ozdemir Can Kara
In recent years, the pervasiveness of colorectal cancer has risen significantly over the world and has become the second leading cause of cancer-related deaths. Current literature reveals that morphological characteristics of a tumor, such as its shape and texture, can help clinicians for early detection and tumor classification. Similarly, the change in the tumor’s modulus of elasticity can be correlated to the cancer stage. However, limitations of existing vision-based and qualitative diagnostic technologies result in an early detection miss rate of 27% for serrated polyps and 34% for flat adenomas. Thus, early detection of cancerous polyps with high sensitivity and reliability becomes imperative to reduce the mortality risk and increase the treatment options. In this study, we propose a framework consisting of a novel vision-based tactile sensor (VTS) and a complimentary machine learning algorithm to ensure sensitivity and reliability of detection. We thoroughly analyze the effect of hardness of sensor and/or tumor on the performance of a VTS in detecting various colorectal polyps by (i) design and fabrication of different VTS with distinct hardness, (ii) design and additive manufacturing of various type, size, shape, and stiffness of realistic colorectal cancer polyps, (iii) thorough experimental evaluation and interaction force measurements of the fabricated sensors on the simulated cancer tumors and (iv) complimentary machine learning algorithm with an appropriate metrics for evaluation of the detection reliability. We demonstrate that by utilizing VTS and solely using visual feedback, we can reliably and sensitively (96.67%) identify different types and stages of colorectal cancer.