A Novel Framework for Sensitive and Reliable Early Diagnosis, Topographic Mapping, and Stiffness Classification of Colorectal Cancer Polyps

Our long-term goal is to develop a novel soft robotic endoscope with intelligent tactile sensing balloons and complementary machine learning (ML) and computer vision (CV) algorithms to enhance early-stage detection, accurate tumor localization, and treatment stratification of various gastrointestinal (GI) cancers. This robotic framework provides clinicians with (i) a safe and intuitively-steerable soft robotic endoscope to perform precise diagnosis, biopsy, and surgical procedures; (ii) in vivo high-fidelity visual, textural, and stiffness information of the diagnosed anatomy; (iii) in vivo radiation-free quantified topographic mapping and morphological characterization (i.e., shape and texture) of GI polyps using CV algorithms; (iv) intelligent real-time in vivo classification of type and stiffness of detected polyps using ML algorithms; and more importantly (v) quantitative evaluations of tumor response during chemo- and radiation-therapy period via in vivo topographic/stiffness mapping. Considering the 2-year timeline of this collaborative project, in this proposal, we will mainly focus on the design, development, and thorough evaluation of a novel and soft Vision-based Tactile Sensing Balloon (VTSB) with complementary Computer Vision (CV) and Machine Learning (ML) algorithms to perform high-resolution in vivo topographic mapping and stiffness classification of Colorectal Cancer (CRC) polyps. CRC is the leading cause of cancer incidence and mortality worldwide. In 2020, CRC accounted for 1.9 million new cases (i.e., #3 cancer type in ranking) and 935,000 new deaths (i.e., #2 cancer type in ranking). Since survival outcomes differ significantly based on the tumor stage at the time of detection, early detection via colonoscopy has a significant impact on treatment outcomes. Morphological characteristics (i.e., shape and texture) and change in the modulus of elasticity of CRC polyps are well-known to be associated with tumor type and stage. Colonoscopic procedures, therefore, are of paramount importance as they can help in early detection and removal of pre-cancerous polyps. However, state-of-the-art traditional colonoscopic procedures still solely rely on visual 2D/3D images and cannot yet provide the clinicians with in vivo detailed textural and stiffness feedback. These limitations has caused high polyp miss rate (about 20%-30%) as well as heavily subjective and evaluator-dependent tumor identification and classifications. It is our central hypothesis that utilizing the proposed VTSB with complementary ML and CV algorithms, can collectively address the limitations of the state-of-the-art colonoscopic technologies by (1) readily integrating with the existing colonoscopic systems and not changing the current clinical diagnosis workflow, (2) providing highresolution 4D imaging (3D texture mapping + stiffness classification), (3) decreasing polyp miss-rate, and (4) enhancing in vivo polyps’ type and stage classification. The proposed contribution is significant, high impact, and innovative and our goal is to demonstrate that it can significantly improve the current diagnosis procedures and shift the current clinical paradigm.