Overview / Usage

Our project aims to develop a deep learning-based system for the prediction of Parkinson's disease, a neurodegenerative disorder that affects millions of people worldwide. It causes unintended or uncontrollable movements, such as shaking, stiffness, and difficulty with balance and coordination. Early diagnosis is crucial for better disease management. Our system utilizes a ResNet-34 convolutional neural network to analyze spiral and wave images and achieves an accuracy of 95%. Healthcare professionals can use this project to diagnose Parkinson's disease accurately and efficiently, leading to better treatment and patient outcomes.

Methodology / Approach

Our methodology involved the use of deep learning technology to develop a predictive model for Parkinson's disease. We utilized the ResNet-34 convolutional neural network architecture for image classification, which has shown excellent performance in similar applications.

To improve model performance, we implemented data augmentation techniques such as random rotations, zooms, and flips to increase the diversity of the image dataset. We also used the transfer learning approach, where the ResNet-34 model was pre-trained on a large image dataset and fine-tuned on our specific task.

We optimized the model by unfreezing the layers and using a cyclic learning rate scheduler that allowed us to gradually increase the learning rate for the first few epochs and then decrease it for the remainder of training. We used the fastai deep learning library to build and train our model.

Our approach resulted in a highly accurate model with a 95% accuracy rate for Parkinson's disease prediction using spiral and wave images. The methodology used in this project can be extended to other medical imaging tasks for the accurate diagnosis of various diseases.

Technologies Used

• Python programming language
• PyTorch deep learning framework
• Fastai library
• Intel one API jupyter notebook
• PIL for image

Repository

https://github.com/Thanush-Prabhu/Parkinson-Disease-Prediction-using-Resnet-34