Platelets Image Classification Through Data Augmentation: A Comparative Study of Traditional Imaging Augmentation and...

Area of Science:

• Medical imaging
• Computational pathology
• Artificial intelligence in healthcare

Background:

• Platelets are vital biomarkers for disease diagnosis and treatment guidance.
• Accurate platelet identification and classification are crucial for clinical applications.
• Limited availability of diverse platelet image datasets hinders deep learning model development.

Purpose of the Study:

• To develop a synthetic platelet image database using Generative Adversarial Networks (GANs).
• To evaluate the effectiveness of GAN-generated data compared to traditional augmentation techniques.
• To assess the performance of various deep learning models on different dataset sizes and augmentation strategies.

Main Methods:

• Generated synthetic platelet images using Wasserstein GAN with Gradient Penalty (WGAN-GP).
• Expanded an initial dataset of 71 images to 141 (Level 1) and 1463 (Level 2) using traditional augmentation.
• Evaluated eight pre-trained deep learning models and two custom CNNs on augmented and synthetic datasets.
• Measured performance using accuracy, precision, recall, and F1-score.

Main Results:

• Deep learning models achieved high performance on extensively augmented data (Level 2), with InceptionV3 and InceptionResNetV2 reaching 99% accuracy.
• DenseNet201 achieved 98% accuracy on Level 2 data.
• GAN-augmented data further boosted DenseNet's performance, indicating its utility in enhancing platelet classification.

Conclusions:

• GAN-based data augmentation shows significant potential for improving platelet classification accuracy.
• Combining traditional and GAN-based augmentation strategies offers a robust approach for medical imaging tasks.
• Synthetic data generation is a promising solution for overcoming data limitations in developing diagnostic AI tools.