[Paper Review] Classification of Alzheimer's Disease Using the Convolutional Neural Network (CNN) with Transfer Learning and Weighted Loss

Muhammad Wildan Oktavian, Novanto Yudistira+1 more|arXiv (Cornell University)|Jul 4, 2022

Public Health and Nutrition22 citations

TL;DR

The paper uses a ResNet-18 CNN with transfer learning from ImageNet, weighted loss, and Mish activation to classify ADNI MRI data into AD, MCI, and CN, achieving 88.3% accuracy (versus 69.1% baseline).

ABSTRACT

Alzheimer's disease is a progressive neurodegenerative disorder that gradually deprives the patient of cognitive function and can end in death. With the advancement of technology today, it is possible to detect Alzheimer's disease through Magnetic Resonance Imaging (MRI) scans. So that MRI is the technique most often used for the diagnosis and analysis of the progress of Alzheimer's disease. With this technology, image recognition in the early diagnosis of Alzheimer's disease can be achieved automatically using machine learning. Although machine learning has many advantages, currently the use of deep learning is more widely applied because it has stronger learning capabilities and is more suitable for solving image recognition problems. However, there are still several challenges that must be faced to implement deep learning, such as the need for large datasets, requiring large computing resources, and requiring careful parameter setting to prevent overfitting or underfitting. In responding to the challenge of classifying Alzheimer's disease using deep learning, this study propose the Convolutional Neural Network (CNN) method with the Residual Network 18 Layer (ResNet-18) architecture. To overcome the need for a large and balanced dataset, transfer learning from ImageNet is used and weighting the loss function values so that each class has the same weight. And also in this study conducted an experiment by changing the network activation function to a mish activation function to increase accuracy. From the results of the tests that have been carried out, the accuracy of the model is 88.3 % using transfer learning, weighted loss and the mish activation function. This accuracy value increases from the baseline model which only gets an accuracy of 69.1 %.

Motivation & Objective

• Motivate automatic early detection of Alzheimer's disease using MRI-based image classification.
• Evaluate CNN-based architectures for Alzheimer's classification with limited data and imbalanced classes.
• Investigate the impact of transfer learning, class weighting, and activation functions on model performance.
• Propose a methodology that mitigates data imbalance and leverages pretraining to improve accuracy.

Proposed method

• Adopted a Residual Network (ResNet) architecture, primarily ResNet-18.
• Applied transfer learning from ImageNet to compensate for limited domain-specific data.
• Implemented weighted cross-entropy loss to address class imbalance among AD, MCI, and CN.
• Replaced ReLU with Mish activation in targeted layers to assess performance gains.
• Preprocessed MRI data by skull-stripping and converting 3D scans to 2D slices for CNN input.

Experimental results

Research questions

• RQ1 Can transfer learning from large-scale datasets improve Alzheimer's MRI classification performance on ADNI data?
• RQ2 Does weighting the loss function to address class imbalance improve detection of the minority AD class and the overall multiclass accuracy?
• RQ3 What is the effect of using Mish activation versus ReLU on model accuracy and class-wise metrics in this task?

Key findings

• Transfer learning with weighted loss and Mish activation yielded 88.3% multiclass accuracy on test data.
• Baseline without transfer learning or weighting achieved 69.1% accuracy, indicating substantial gains from the proposed approach.
• Weighted loss improved precision for the AD and CN classes and increased several F1-scores across classes.
• Transfer learning significantly increased accuracy, precision, recall, and F1-score in the reported results.
• The Mish activation function did not further improve accuracy when combined with transfer learning and weighted loss, though it affected some class-wise metrics.