ATTNFNET: feature aware depth-to-pressure translation with cGAN training.

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Artificial Intelligence in Healthcare

Background:

• Bedridden patients are susceptible to pressure injuries due to excessive pressure and shear forces.
• Existing ulcers exacerbate the risk of pressure injuries in vulnerable patients.
• Accurate monitoring of pressure distribution is vital for early detection and prevention.

Purpose of the Study:

• To develop a novel deep learning model for generating pressure distribution maps from single depth images.
• To improve the accuracy of pressure injury prevention strategies through enhanced pressure monitoring.
• To introduce the Attention Feature Network (AttnFnet) for precise pressure mapping.

Main Methods:

• Utilized a self-attention-based deep neural network, AttnFnet.
• Employed Conditional Generative Adversarial Network (cGAN) training for map generation.
• Introduced a mixed-domain SSIML2 loss function combined with adversarial loss.

Main Results:

• AttnFnet demonstrated superior performance compared to existing methods.
• Achieved high accuracy in pressure distribution estimation from single depth images.
• Evaluation metrics included Structural Similarity Index Measure (SSIM) and quality analysis.

Conclusions:

• AttnFnet provides an accurate and effective method for estimating pressure distributions.
• The proposed model aids in identifying high-risk areas for pressure injury prevention.
• Single depth image analysis offers a promising approach for non-invasive pressure monitoring.