Fully Automatic Severity Grading of Lower Limb Arterial Blockage based on Plantar Perfusion Imaging

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Machine Learning in Healthcare

Background:

• Lower limb arterial blockage (LLAB) is a key indicator of peripheral arterial disease (PAD).
• Current diagnostic methods for LLAB can be invasive or costly.
• Plantar perfusion imaging offers a non-invasive approach for assessing LLAB.

Purpose of the Study:

• To evaluate machine learning methods for grading LLAB severity using plantar perfusion images.
• To develop and validate a novel deep learning model, ResNet18-CL, for improved LLAB severity classification.

Main Methods:

• Simulated LLAB by applying five pressure levels to the calf in 20 healthy participants.
• Collected plantar video data and generated remote photoplethysmography (rPPG) images.
• Benchmarked four conventional machine learning algorithms and proposed the ResNet18-CL model.

Main Results:

• ResNet18 demonstrated superior performance over traditional machine learning methods for LLAB grading.
• The proposed ResNet18-CL model achieved significant improvements: 5.98% in accuracy and recall, 6.58% in precision, and 7.36% in F1 score compared to ResNet18.
• The model effectively distinguished between different LLAB severity levels.

Conclusions:

• Camera-based plantar perfusion analysis is a viable method for LLAB assessment.
• The ResNet18-CL model offers a timely, cost-effective, and efficient solution for LLAB severity grading.
• This approach holds promise for early PAD prevention and management.