Accurate detection of tumor boundaries is critical for the success of oncologic surgical intervention. Traditionally, palpation can handover important information for tumor localization based on the tissue mechanical properties, but in Minimally Invasive Surgery no direct access to the tumor for palpation is feasible. For providing a technical analogy, this feasibility-level study focusses on the simplified problem of detection of an inclusion within a homogeneous silicon phantom. We hypothesized that existence of a relatively stiffer inclusion within an elastomer tissue phantom changes vibroacoustic signatures under forced vibration conditions. In comparison with previous studies, in this work the measurement probe was static, and the short-time (1 s) data package analysis targeted at nearly real-time inclusion detection. The inclusion detection problem was cast into a binary classification of the short-time acquired vibroacoustic signals. The method involves a wavelet-based multilayer perceptron neural network (MLP) that is trained in a supervised manner. A micro-electro-mechanical system (MEMS) sensor proximally attached to a solid probe was used to measure the vibroacoustic signals. Phantoms of simulated healthy tissue with stiffer tumor model inclusions were used for experiments and data collection. From the 120 overall number of experiments, 15 % were used as test data to evaluate the performance. The results show inclusion detection F1 score of 75 %, and 77.8 % accuracy related to the confusion matrix, reflecting the model performance on previously unseen data. Performance of the classifier was discussed in terms of various binary classification metrics, and compared with another established classifier, support vector machine (SVM). While the results support the hypothesis of this proof-of-concept study, extensions like improving the electronic system and refining the method with more experiments on biological tissues remain as the future work.
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号