MRI and CT image indexing and retrieval using steerable pyramid transform and local neighborhood difference pattern

Abstract

This paper proposes an effective system for indexing and retrieval of medical images, which uses steerable pyramid transform (SPT) and local neighborhood difference patterns (LNDP). In the first step of the proposed system, a medical image is subdivided into several sub-bands using SPT at different resolutions and orientations. Further, each sub-band is encoded using LNDP in a binary texture pattern to extract more complex texture features. Finally, LNDP histograms are computed for each SPT sub-band. Thereafter, all the sub-band histograms are combined to form a single histogram to produce the required feature vector. The proposed SPT-LNDP technique represents a medical image in an efficient way that contains many sources of information from various scales and orientations. Two open-source medical image datasets namely OASIS- MRI and VIA/I-ELCAP-CT images of human brain were considered for efficient evaluation of the proposed method. To validate the efficacy of the proposed method two parameters namely Average Retrieval Precision (ARP) and Average Retrieval Rate (ARR) were used. A detailed comparative study of the proposed method with other popular methods, which are described in this study, are also presented in this work. The results show improvement over other existing methods considered in this study.