Multi-Stream Attention-Based Convolutional Neural Network for Medical Imaging
V. V. Satyanarayana Tallapragada, B. Bhaskar Reddy, A. Swetha Rani, P Sree Lakshmi, Hedayath Basha Shaik · Advances in medical diagnosis, treatment, and care (AMDTC) book series · 2025
In the recent past medical imaging has emerged as a tool for better diagnosis of patients. The diagnosis mechanism includes Non-invasive methods and further treatment of various diseases viz., pneumonia cancer and other neurological disorders. To address the challenges in the existing methods, this chapter introduces a novel Multi-Stream Attention-Based Convolutional Neural Network (MSA-CNN) technique which is a deep learning architecture. The main component of the proposed architecture is an Attentional mechanism that dynamically assigns prominence to the features of each modality, concentrating on the most relevant regions of the images that are of interest, to improve the models interpretability. Further, a Noise-Aware Loss Function is introduced to handle noise in the images to enhance model performance. Results show that the proposed technique is observed to have an accuracy of 95.7% on the Chest X-ray dataset, outperforming the existing benchmarking techniques to have 89.3% using AlexNet, 92.4% using ResNet-18 respectively.