The Design of a Biomimetic Whisker-based System for Clinical Gastrointestinal Diseases Screening

Abstract

The prevalence of gastrointestinal diseases, manifesting as many broad and confusing symptoms such as malabsorption, malnutrition, and even neurologic dysfunction[1], has become a socio- economic burden for present-day society. Extensive clinical studies illustrate that early screening for gastrointestinal diseases is of vital importance for reducing mortality and improving life expectancy. The conventional screening method is endoscopy, featured by high sensitivity in diagnoses of mucosal diseases[2]. Its minimally invasive nature makes it is extensively carried out in hospitals. However, the highly confined and challenging working conditions including darkness, glare, reflection, bleeding caused by biopsy, blur, and defocus can all prevent optical sensors from working optimally, even lead to missing diagnosis[3]. Meanwhile, the diagnosis bias induced by endoscopist differences has a significant effect [4]. In addition, the tactile sensory modality is desired in surgical practice, which is the foundation of efficient force feedback in minimally invasive surgery. Inspired by the sensory mechanism of rat whisker, which shows great potential in texture discrimination and distance estimation tasks[5], in this paper, a biomimetic artificial whisker system is proposed for clinical gastrointestinal disease screening. Unlike optical sensing methods, this tactile sensing is a strong complement to the conventional vision-based methods, especially in situations where optical devices cannot work well, a totally lightless environment for instance. Utilizing the analog front end (AFE) technology, the system inherent noise is minimized, which suppresses detection error in experiments. An end-to-end deep learning algorithm is proposed to provide diagnostic outcomes assisting clinical decisions and avoiding subjective bias. A medical phantom-based pilot study was conducted to demonstrate the detection accuracy of the proposed with three common tissue structures, which are normal tissue, ulcerative colitis, and ulcerative cancer. The results have shown the proposed can reach 100% in distinguishing the different types of tissues, and demonstrated the potential of this novel system for advancing endoscopy.