Integrating CNN along with FAST descriptor for accurate retrieval of medical images with reduced error probability

The size of medical image repositories is continuously growing due to the widespread use of digital imaging data in hospitals. This overlays the way for more medical records to be stored in the future. An effective image retrieval system must be designed to make retrieving medical images from datasets as simple as possible. Using diverse feature extraction procedures, a number of researchers have created several picture retrieval frameworks. However, a semantic gap caused by poor extraction of low and high-level features is a fundamental problem in traditional image retrieval frameworks. As a result, during the construction of a retrieval framework, an effective feature extraction technique must be included for proper extraction of both level characteristics. The present research aims at designing modified Convolutional Neural Network(CNN) for the effective retrieval of medical images. The proposed process is performed using two models such as training and testing model. In the training phase, the features are learned using Features from Accelerated Segment Test with CNN (FAST-CNN) and stored in the database. Subsequently, in the testing process, a query image is retrieved from the dataset based on the feature matching process using Minkowski distance. The performance of the proposed retrieval framework is tested on three medical datasets using some of the metrics such as accuracy, sensitivity, precision, and F1 score. Using the proposed retrieval framework, accurate retrieval with a lesser error rate is achieved and the accuracy reached using this proposed retrieval framework is around 94%.