Integrating 3D Model Representation for an Accurate Non-Invasive Assessment of Pressure Injuries with Deep Learning

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Wound Care Technology

Background:

• Pressure injuries are a significant health concern, particularly for elderly and disabled individuals.
• Current invasive methods for quantitative assessment cause pain, discomfort, and infection risks.
• 2D imaging methods lack the depth and shape information crucial for accurate pressure injury measurement.

Purpose of the Study:

• To develop a non-intrusive system for accurate pressure injury assessment.
• To overcome limitations of traditional 2D imaging by incorporating 3D data.
• To provide caregivers with reliable wound measurements for medical assessment and healing evaluation.

Main Methods:

• An end-to-end system utilizing a single 2D image and a 3D mesh acquired via Structure Sensor.
• A Mask RCNN model for segmenting external wound boundaries from 2D images.
• A 3D mesh segmentation block that integrates 2D segmentation to derive real-world measurements (depth, area, volume).

Main Results:

• The proposed framework achieves refined wound segmentation with 87% precision.
• The system successfully generates reliable real-world measurements of pressure injuries.
• Accurate 3D data extraction from combined 2D and 3D inputs.

Conclusions:

• The developed non-intrusive system offers a valuable tool for pressure injury assessment.
• The method provides accurate segmentation and quantitative measurements, improving patient comfort and care.
• This technology aids in medical assessment and monitoring the healing progress of pressure injuries.