Impulse: The Premier Undergraduate Neuroscience Journal (Sep 2024)

Employing a Convolutional Neural Network for multi-class classification of Alzheimer’s Disease through MRI scans

  • Kashish Kapoor,
  • Virginia Fernandez,
  • Amparo Guemes

Abstract

Read online

Alzheimer’s Disease (AD) is characterized by progressive cognitive decline, and the prevalence of the disease continues to rise as the population ages. With over 10 million new cases yearly, there is still no established way to quickly and accurately classify the stage a patient with Alzheimer’s has progressed to. Inaccurate classification leads to delayed or incorrect treatment, financial implications, and emotional distress, ultimately severely impacting both the patient and their family. This paper proposes a novel deep learning Convolutional Neural Network (CNN) model to assist in the classification of a patient's AD. The model development began with a dataset of over 6,000 pre-labeled Magnetic Resonance Imaging (MRI) scans categorized into four stages: normal, very-mild, mild, and moderate AD. Using transfer learning with InceptionV3, the framework analyzes class proportions and fine-tunes a CNN to leverage pre-trained features. The CNN then undergoes rigorous training with early stopping based on validation Area Under Curve (AUC). Upon completing the training, a comprehensive model evaluation is conducted, encompassing metrics such as AUC and confusion matrices. Using these methods, three different mini-models are constructed, all with different training parameters and number of epochs. Conclusively, the successful models are saved, and an array of evaluation metrics such as accuracy, precision, F1 score, and recall are then used to analyze the models' results. The results show that the proposed method achieves 95.0% or more across all metrics in seconds, demonstrating a promising classification performance that can greatly assist doctors in the classification of AD, helping achieve not only higher diagnostic accuracy but also allowing for faster diagnosis. The results of this model demonstrate significant potential to enhance patient care and highlight the impact of Artificial Intelligence (AI) on AD diagnosis. With the ability to classify MRIs with 20 percent greater accuracy and reduce diagnostic time from months to seconds, the promise of AI in not only AD diagnosis, but the whole of healthcare will enable us to take steps toward a better world. Envisioning a future where the healthcare system can prioritize patient wants rather than just needs. Abbreviations: AD - Alzheimer’s Disease; AI - Artificial Intelligence; AUC – Area Under Curve; CNN - Convolutional Neural Network; RNN - Recurrent Neural Networks; ROC - Receiver Operating Characteristic

Keywords