Alzheimer’s diagnosis transformation: Evaluation of the effect of CLAHE on the effectiveness of EfficientNet architecture in MRI image classification

Abstract

Alzheimer’s disease is a global health challenge with an increasing number of cases, particularly in developing countries such as Indonesia. Early diagnosis is crucial to slowing the progression of this disease. This study evaluates the impact of Contrast Limited Adaptive Histogram Equalization (CLAHE) on the quality of Magnetic resonance imaging (MRI) images to enhance the performance of deep learning models, namely EfficientNet-B3 and EfficientNetV2-B3, in classifying Alzheimer’s disease into four categories: Moderate Demented, Mild Demented, Very Mild Demented, and Non-Demented. CLAHE is applied to enhance the local contrast of MRI images, making important features more visible. The results show that the EfficientNetV2-B3 model with CLAHE achieves 99% precision, 99% F1-score, and 98% accuracy, while EfficientNet-B3 with CLAHE also shows significant improvements compared to models without preprocessing and those using Histogram Equalization (HE). CLAHE has proven not only to improve accuracy but also to stabilize classification, particularly for minority classes such as Moderate Demented, which are difficult to detect using conventional methods. This study highlights the importance of CLAHE as part of the development of AI-based diagnostic tools for Alzheimer’s, especially in clinical environments with limited resources. The main contribution of this research is demonstrating how CLAHE, when integrated with modern architectures such as EfficientNet-B3 and EfficientNetV2-B3, not only enhances the model’s sensitivity to critical features in MRI data but also establishes a new approach to improving classification outcomes in real-world scenarios with resource constraints.