Quantitative 3-D shape features based tumor identification in the fog computing architecture

Abstract

For the purpose of accurate diagnosis and early treatment for cancers, the classification and identification of different tumors is the key problem of computer-aided diagnosis system. In this paper, an improved semi-supervised tumor identification method is proposed, which takes advantage of the Fuzzy c-means clustering algorithm and offers a pathological degree tree based on ten three-dimentional (3-D) and two dimentional (2-D) tumor features. In addition, a great deal of complicated data processing is distributed in the fog computing architecture. First, we carry out the segmentation of tumors by using FRFCM algorithm, and complete the 3-D modeling. Then, the pathological shape features of 3-D and 2-D tumors are extracted from modeling, for constructing a group of feature vector. Finally, based on the landmark information of labeled samples provided by standard database and experts, we realize an improved semi-supervised FCM clustering to guide the tumor identification. The experiments are conducted by using medical CT scans of 143 patients including 452 tumors. Overall, the best average identification accuracy of $$94.6\%$$ has been recorded for this proposed method, the ability of machine learning to recognize the benign, malignant and false-positive tumors is improved effectively under imbalanced data sets.