An autonomous and intelligent hybrid CNN-RNN-LSTM based approach for the detection and classification of abnormalities in brain

This article shows the identification and classification of abnormalities present in the brain. MRI test is typically conducted to detect abnormalities in the brain, but the test gives multiple information from a single image, which becomes highly exhaustive and challenging. Such types of issues can be resolved by considering the multi-mark classification, i.e., allocating multiple images with more than one mark. The marks are represented in terms of brain abnormalities. The six abnormalities of the brain are taken into account, namely: infract, hemorrhage, ring-enhancing lesion, granuloma, meningitis, and encephalitis. In order to detect and classify the abnormalities, convolutional neural networks (CNN) and recurrent neural networks (RNN) are used. CNN is used to extract the important features of the input signal based on a channel-wise model, while RNN is used to classify the various abnormalities with dependency parameters. RNN is executed with long short-term memory (LSTM) in order to prevent gradient failure. Performance parameters like accuracy, precision, probability of occurrence (POC), and mean square error (MSE) are used to avoid boundary conditions and classify abnormalities. Still, it is observed that individual applications of CNN and RNN-based LSTM for the detection and classification of abnormalities provide inappropriate performance parameters and involve huge mathematics. In order to resolve such issues, the best features of CNN and RNN-based LSTM methods have been extracted and developed in the hybrid intelligent controller. The hybrid approach provides improved and better performance parameters for the appropriate image classification of abnormalities in comparison to individual CNN, RNN, RNN-based LSTM, and other existing methods. The effectiveness and testing of the proposed hybrid approach are being tested on samples of 1000 collected data from the standard source. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2023.