Design And Implementation Of Pulmonary Nodules Similar Images Retrieval System

Lung cancer is one of the main diseases that threaten human health, but its early diagnosis and treatment can significantly improve the cure rate. In the early stage of lung cancer, it is usually a solitary pulmonary nodule. The discovery and diagnosis of pulmonary nodules by the doctor through reading the lung CT(Tomography Computed) scan images, which the workload is big and strong subjectivity. With the help of computer image processing and analysis technology, and through the detection of pulmonary nodules and similar nodules search, it can assist physicians according to the characteristics, pathology and diagnosis information of pulmonary nodules which is searched, to diagnose and analyze the current pulmonary nodules.The similar image search framework designed in this paper includes three parts: localization of pulmonary nodules and ROI extraction, feature extraction and similarity ranking. Lung nodule detection and ROI extraction module mainly include: two-dimensional Otsu segmentation algorithm on the thoracic, to remove elongated tissue of the lung wall noise, extraction of lung parenchyma, enhanced multi-scale Hessian algorithm enhancements to spherical structure, and extraction of pulmonary nodules ROI. In the aspect of feature extraction, using currently popular deep convolutional neural network technology, to artificial markers similar image samples to train the Siamese network of "twin towers" structure, and automatic extraction features, which it solves the using artificial experience to define the characteristics of calculating formula of difficulties, and avoids the problem that the traditional feature extraction methods rely on accurate segmentation of nodules. Pulmonary nodules similarity ranking module mainly used LambdaMART model to predict candidate pulmonary nodules ratings. The input of the LambdaMART model is the difference between the characteristics as a new sample characteristics in ROI image library. Eventually, we will were able to get a set of images similar with the image searched, and it gives the similarity score.The similar image searching method of the paper make experiments about extracting the subject pulmonary nodule ROI in LIDC-IDRI(Lung Image Database Consortium and Image Database Resource Initiative) image library, pulmonary nodules ROI feature extraction and similarity searching pulmonary nodules. The results show that this method can effectively extract the ROI characteristics of pulmonary nodules and can get satisfactory pulmonary nodules similar searching results.