IoT-based Stomach abnormality detection via hybrid MDCNN-Bi-LSTM architecture with statistical and texture features

Stomach abnormalities pose significant health concerns, ranging from minor digestive issues to severe conditions. The emergence of deep learning methods offers a promising solution to this problem. However, due to the risk of non-optimal hyperparameters affects its performance. To address this concern, this research proposes a Healthcare Internet of Things (IoT)-Based Stomach Abnormality Detection through Iris Image (HIoT-SADII). The SADI process begins with data collection through an IoT architecture. The data are preprocessed using a Gaussian filter. A Modified Mean with Niblacks' Threshold-based Deep Joint Segmentation (MMNT-DJS) is suggested for segmentation that separates the region of interest from background. Subsequently, features such shape features, statistical features, and Modified Local Gabor Transitional Pattern (MLGTP) are extracted from segmented images. Lastly, a hybrid deep learning approach that incorporates Bi-Directional Long Short-Term Memory (Bi-LSTM) and Block-Wise Modified Dropout in Convolutional Neural Network (BMDCNN) is proposed for detection. The proposed model is trained with extracted features to determine final outcome as normal or abnormal based on averaging both models' outcomes. Experimental findings demonstrate that the suggested model achieves a detection accuracy and F-measure of 0.950 and 0.927, respectively.