Deep learning based approach with efficientNet and SE block attention mechanism for multiclass alzheimer's disease detection

Abstract

Alzheimer's disease can be detected beforehand using brain MRI images. However, this situation requires a specialist in the field and is a very open area to human error. In addition, it takes a long time for specialist doctors to analyze images and make decisions by diagnosing diseases. In this study, it is aimed to detect Alzheimer's disease quickly and with high accuracy by using deep learning architectures. In addition, with the use of the proposed study in hospitals, it can play an important role in reducing false treatments by supporting doctors in disease detection. In the study, open source licensed brain MR images dataset obtained from Kaggle was used. The dataset is divided into 3 subgroups as training, validation and testing. Within the scope of the work, where EfficientNet-B0 architecture and SE block attention mechanisms were used, a model specific to the data set was developed and used. The results of the studies were compared with other studies using the same data set and detailed results were given in the study. Studies have shown that Brain MRI images can be successfully classified with an accuracy of 0.9903. The findings suggested that the architecture created might help reduce human error in diagnosing Alzheimer's disease.