Attention-based CNN for CT-Kidney Classification

We constructed an attention-based convolutional neural network (CNN) model structure, and it was applied to classify CT-kidney. The concept of attention mechanism and its application in refining con-volution-extracted feature map is elaborated. The addictive attention mechanism is adopted to allocate more attention weight to more distinguish region on feature map, and lead to improvement in classification accuracy. Two experiments were conducted to verify our model structure. In the first experiment, the efficiency of attention-based CNN was validated as it reaches fully convergence within 8 epochs and achieve 100% accuracy on test set. The second experiment verify its improved performance, lightweight and easily train-able architecture in comparison with CNN and Resnet50 on Alzheimer's MRI dataset. In the future, this approach could potentially be employed to address challenges associated with CT radiography classification.