Towards the analysis of dynamic medical thermography : a framework for registration, segmentation and component analysis

Abstract

Maintaining foot health is vital to overall well-being, especially for individuals with diabetes, who are at an elevated risk of developing lower-limb complications due to neuropathy and poor circulation. These conditions heighten the risk of infections, foot ulcers, and, in severe cases, amputations. Monitoring foot temperature offers a non-invasive and effective means of early detection for such complications. Shifts in temperature can indicate underlying issues: reduced temperatures may point to circulatory disorders such as peripheral arterial disease (PAD), while increased temperatures often suggest inflammation or ulceration, which can quickly worsen if left untreated. The thesis explores the application of dynamic thermal imaging for foot health monitoring. Thermography, a non-contact technique that detects infrared radiation emitted from the skin, was employed to visualise and analyse temperature distribution across the foot. Unlike traditional static imaging, this study utilises video-based thermal data to capture temporal changes in temperature distribution. However, incorporating a temporal dimension introduces additional challenges, notably subject movement during video capture. To address this, the study proposes an automated registration method capable of aligning the foot in dynamic thermal sequences without user intervention. A VoxelMorph-trained SynthMorph network was applied to linearly pre-registered videos, and results demonstrate that the method achieves high registration accuracy. The next stage in the processing pipeline involves isolating the foot region from the background in the thermal data. This was achieved using the Segment Anything Model (SAM), which delivered 100% segmentation success with minimal manual input. The final step involves analysing the temperature dynamics within segmented foot region. Due to the complex interplay of physical and environmental factors influencing thermal data, interpretation can be challenging. To address this, Principal Component Analysis (PCA) was used to decompose the temperature dynamics. The findings show that PCA effectively distinguishes between different underlying processes. Overall, the study demonstrates the viability of dynamic thermal imaging for continuous foot health monitoring and highlights its potential as a preventive healthcare tool. Thermography is shown to be reliable, efficient, and accessible method for tracking foot health, paving the way for automated solutions in preventative care. By providing early warning signals, dynamic thermal imaging may improve patient outcomes, lower healthcare costs, and enhance overall quality of life.