부산대학교 도서관

An endoscopic tactile robotic capsule, embedding miniaturized MEMS force sensors, is presented. The capsule is conceived to provide automatic palpation of non-polypoid colorectal tumours during colonoscopy, since it is characterized by high degree of dysplasia, higher invasiveness and lower detection rates with respect to polyps. A first test was performed employing a silicone phantom that embedded inclusions with variable hardness and curvature. A hardness-based classification was implemented, demonstrating detection robustness to curvature variation. By comparing a set of supervised classification algorithms, a weighted 3-nearest neighbor classifier was selected. A bias force normalization model was introduced in order to make different acquisition sets consistent. Parameters of this model were chosen through a particle swarm optimization method. Additionally, an ex-vivo test was performed to assess the capsule detection performance when magnetically-driven along a colonic tissue. Lumps were identified as voltage peaks with a prominence depending on the total magnetic force applied to the capsule. Accuracy of 94 % in hardness classification was achieved, while a 100 % accuracy is obtained for the lump detection within a tolerance of 5 mm from the central path described by the capsule. In real application scenario, we foresee our device aiding physicians to detect tumorous tissues.