Adaptive Augmentation of Medical Data Using Independently Conditional Variational Auto-Encoders.

Area of Science:

• Medical Imaging
• Machine Learning
• Data Science

Background:

• Deep supervised learning models require extensive labeled data for training.
• Acquiring and labeling clinical data is costly, resulting in small medical datasets.
• This limitation hinders the development of robust deep learning models in healthcare.

Purpose of the Study:

• To propose a variational generative model for effective data augmentation in medical imaging.
• To address the challenge of limited labeled data in clinical settings.
• To improve the performance of deep learning models using synthesized data.

Main Methods:

• Developed a variational generative model learning conditional probability distributions of image data.
• Utilized the generative model to synthesize new images for data augmentation.
• Evaluated the approach on clinical datasets including spine ultrasound and brain MRI.

Main Results:

• Improved accuracy for spine image classification: 92% (residual model) vs. 83% (conventional).
• Enhanced Dice coefficient for brain tumor segmentation: 88% (U-net) vs. 83% (conventional).
• Demonstrated significant performance gains compared to traditional training methods.

Conclusions:

• The proposed variational generative model effectively synthesizes data for augmentation.
• This approach alleviates the need for large labeled datasets in medical deep learning.
• The method shows promise for improving diagnostic accuracy and segmentation tasks in clinical practice.