Improving Parkinson's Disease Detection with Hybrid Deep Learning and Machine Learning
Author Information
Author(s): Khedimi Madjda, Zhang Tao, Merzougui Hanine, Zhao Xin, Geng Yanzhang, Djaroudib Khamsa, Lorenz Pascal, Zhang Teng Grace, Cheung Jason Pui Yin, Zeng Tianjiao, Meng Nan
Primary Institution: Department of Electrical and Information Engineering, University of Tianjin, Tianjin, China; Department of Computer Science, University of Batna 2, Batna, Algeria; Department of Computer Science, IUT of Colmar University of Haute Alsace, Colmar, France.
Hypothesis
How do different hybrid deep learning and machine learning models respond to various data augmentation techniques when applied to spiral drawing data for Parkinson's Disease detection?
Conclusion
The study found that data augmentation has a variable impact on different hybrid models, with some models benefiting significantly while others experienced performance degradation.
Supporting Evidence
- ViT with KNN achieved 96.77% accuracy on unaugmented data.
- ResNet-50 with Logistic Regression improved to 93.55% accuracy with data augmentation.
- The study emphasizes the need for appropriate augmentation techniques in medical image analysis.
- Hybrid models respond differently to augmentation, necessitating careful selection of strategies.
- Data augmentation is critical for enhancing model generalization in small datasets.
Takeaway
This study looked at how well different computer models can detect Parkinson's Disease by analyzing spiral drawings, and found that using extra training data can help some models work better.
Methodology
The study used a hybrid approach combining deep learning and machine learning techniques to analyze spiral drawings for Parkinson's Disease detection, employing various data augmentation strategies.
Potential Biases
The dataset may not be fully representative of the diverse population of Parkinson's Disease patients, which could affect model performance.
Limitations
The study faced limitations due to the small dataset size, which may lead to overfitting and limit generalizability, and the lack of external validation.
Participant Demographics
The dataset includes spiral drawings from individuals diagnosed with Parkinson's Disease and healthy controls, balanced across both groups.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website