Registration of thermographic video for dynamic temperature analysis in humans

Abstract

The use of infrared thermography in medical applications has increased in popularity in recent years. It facilitates the detection and examination of skin thermal signatures, under both normal and abnormal conditions. Thermography has been employed in numerous biomedical fields, including breast cancer detection, cutaneous temperature monitoring during exercise and the analysis of normative temperature patterns. Thermal imaging may be dynamic or static in nature. Using static thermography, the steady state conditions and spatial distributions of the thermal patterns within a target are analysed at a particular instant, usually following an acclimatisation period. In contrast, via dynamic thermography, both spatial and temporal variations are considered, making the acquired data more informative. However, issues including involuntary target movement and the dynamic temperature changes undergone by the target need to be considered. Video registration was opted for in this work. Four steps constitute the registration process. The Speeded-Up Robust Features (SURF) detector was utilised in the feature detection stage. Matching features between images were then found based on the sum of squared differences (SSD) error, following which an affine geometric transformation was computed to adequately map the images in consideration. Bilinear interpolation was then utilised to calculate pixel values in non-integer coordinates. Two video registration methods were proposed in this work to address the primary issues associated with dynamic thermography. Data was gathered from nine participants for the testing of these methods. Following implementation, their performance was assessed both qualitatively and quantitatively, and a two-sample ttest was applied to verify that the difference between the mean errors per method was statistically significant. Dynamic temperature analysis was also carried out on the extracted temperature data in both the time and frequency domains, where cyclic patterns having different frequencies and magnitudes were observed across all participants. Such behaviour has not been documented in literature thus far, which implies that the biological significance of these patterns is yet to be determined.