DATDL-DCNN-BiLSTM: Dual Attention Temporal Difference Learning Based Distributed Deep Learning Model for Brain Tumor Detection

Abstract

The brain tumor (BT) is a critical disorder related to neurons characterized through the prolif-eration growth. The survival rates associated with this disease are steadily declining, primarily due to insufficient early detection and precise diagnosis of BT. The manual inspection often suffers from subjectivity, low accuracy, and inefficiency due to complex tumor shapes. To identify and handle the large sized datasets along with capturing the more subtle tumor fea-tures, this study proposes a model named socio-swarm intelligence optimizer (SSIO) enabled dual attention temporal difference learning with distributed convolutional neural network and bidirectional long short-term memory (SSIO-DATDL-DCNN-BiLSTM) framework for iden-tification of BT, focusing on the integration of advanced optimization techniques. The dual attention-AlexNet layer efficient statistical triangular ResNet (DA-ALESTR) provides detailed localization of relevant features. The LayerCAM (layer-wise class activation map) and AlexNet features in the model add more deliberation by capturing high-level semantic patterns and enhancing interpretability. The SSIO optimizer increase the interpretation through adjusting the factors. The results of the proposed SSIO-DATDL-DCNN-BiLSTM demonstrates higher scores on accuracy, recall, F1-score, and precision with 99.56%, 98.27%, 96.80%, and 99.79% with training data, and with k-fold 96.95%, 96.96%, 96.77% and 97.15% using BraTS 2018 dataset.