AlziFusionNet: Alzheimer's disease detection Using transfer learning-based hybrid Deep learning model.

Abstract

Neurological ailments such as Alzheimer's Disease (AD) drastically affect brain function and cognitive capacity. With the growing incidence of AD, early and accurate diagnosis is essential for effective treatment and patient management. There is no cure for this disease and a lot of care by caregivers is required. As the disease progresses, it also hampers the family relationship as there is not much awareness among people. Early detection of this disease will help in slow progression along with various therapies, and psychological and cognitive tests. Alzheimer’s detection at an early stage may change the life of the patient, as early detection of the disease with the help of drugs and therapy can slow the progression of the disease. There are various stages of this neurological problem. MRI Magnetic Resonance Imaging and CT Computed Tomography scan can be utilized to study multiple stages. A data set from Kaggel consists of 6400 MRI images and is divided into four classes. Image normalization technique is applied along with the data pipeline to optimize the performance. various parameter settings are done to get the highest accuracy. AlziFusionNet is proposed to distinguish Alzheimer affected brain and normal brain from MRI. it is based on a transfer learning-based combination, applying global average pooling and batchnormalization, will lessen the overfitting and the model is computationally efficient. AlziFusionNet demonstrated encouraging results and remarkable performance. It proves it is a benchmarking technique. This AlziFusionNet achieved a testing accuracy of  99.87% and a training accuracy of 99.90% with precision, recall, and F1 score between 1.00 and 0.98. The model is highly optimized and demonstrates strong generalization performance, even for classes with limited data.