Design and development of case based retrieval system for interstitial lung diseases

Abstract

Lung diseases are the third leading cause of death worldwide. According to WHO report, more than 3.23 million people died in 2019 from lung diseases. The advancements in the medical field have made it possible to identify these diseases using CAD systems. In literature, lots of CBMIR systems are developed to retrieve similar tissue patterns but very few case- based retrieval systems exist. CBMIR systems utilize classification modules and retrieve images of tissue patterns which have less clinical utility. In this quest, case-based retrieval systems are utilized. The present study has developed a framework for the case-based retrieval system for the retrieval of similar ILD cases by fusing derived imaging features and clinical parameters of the ILD dataset MedGIFT. The proposed framework comprises three modules: segmentation of the lung field region, feature extraction and selection of patches and medical image retrieval systems. The performance of the proposed Lung-Seg-IIFCM segmentation algorithm is measured using a dice similarity coefficient and Jaccard similarity coefficient and obtained values are 0.951 and 0.924 respectively. After lung segmentation, patches are generated using overlapping and non- overlapping approaches. The performance of both of the approaches is tested using different classifiers. Results show SVM-RBF efficiently differentiates the lung tissue patterns with an F1-score of 99.47% and an accuracy of 99.63%. SVM-RBF is further utilized to predict the tissue patterns and count of the tissue patterns of each class are used as derived imaging features and fused with clinical parameters. Later, by using multimodal fusion criteria five similar ILD cases are retrieved to understand more about the query case. The performance of the proposed case-based retrieval system is measured in terms of precision metric having a value of 94.34% for Pat1 and 78.49% for Pat5.