Related Experiment Video
Updated: Jun 2, 2025
05:06
Non-Invasive Visualization of Nailbed Microvascular Morphology in Mice Using Capillaroscopy
Published on: February 28, 2025
188
Abnormality detection in nailfold capillary images using deep learning with EfficientNet and cascade transfer
Mona Ebadi Jalal1, Omar S Emam2, Cristián Castillo-Olea3
1Hive AI Innovation Studio, Department of Computer Science and Engineering, University of Louisville, Louisville, KY, 40292, USA. mona.ebadijalal@louisville.edu.
Scientific Reports
|January 15, 2025
Summary
A new deep learning model using cascade transfer learning on Nailfold Capillaroscopy (NFC) images achieved perfect accuracy in distinguishing normal from abnormal cases. This automated tool shows promise for early disease detection and improved patient care.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Microvascular Medicine
Background:
- Nailfold Capillaroscopy (NFC) detects microvascular changes indicative of systemic diseases like diabetes and systemic sclerosis.
- Current manual NFC evaluation is time-intensive, subjective, and relies on expert interpretation.
- Automated analysis of NFC images can enhance diagnostic efficiency and accuracy.
Purpose of the Study:
- To develop and validate a deep learning framework for automated classification of Nailfold Capillaroscopy images.
- To assess the performance of a cascade transfer learning approach using EfficientNet-B0 for differentiating normal from abnormal NFC findings.
- To establish a robust automated clinical screening tool for early detection of microvascular abnormalities.
Main Methods:
- A dataset of 225 NFC images was utilized, with normal cases comprising 6% of the data.
- A cascade transfer learning framework based on EfficientNet-B0 was implemented.
- The model was pre-trained on ImageNet and fine-tuned using domain-specific NFC data.
Main Results:
- The proposed cascade transfer learning model achieved perfect performance metrics: 1.00 for accuracy, precision, recall, and F1 score, and 1.00 for ROC_AUC.
- This significantly outperformed single transfer learning models (0.67 accuracy, 0.83 ROC_AUC) and a CNN-based cascade model (0.67 accuracy, 0.83 ROC_AUC).
- The model demonstrated high efficacy in distinguishing between normal and abnormal NFC images.
Conclusions:
- Deep learning, specifically cascade transfer learning with EfficientNet-B0, offers a viable and highly effective automated approach for NFC image analysis.
- This automated framework has the potential to significantly improve early disease detection and preventive healthcare strategies.
- The study highlights the capability of AI to enhance diagnostic tools, leading to better patient outcomes and quality of life.
