<p>Infrared thermal imaging has emerged as a tool in medical research and diagnostics, offering a potential high throughput AI-powered method to analyze skin temperature patterns and find physiological conditions. This study focused on the research application of digital thermal imaging to detect hot flush events in primates. A convolutional neural network (CNN) was employed to define facial boundaries and measure facial temperatures. To enhance boundary detection, edge detection methods were applied as a preprocessing step, with their outputs serving as inputs to the CNN model. These methods were evaluated and compared on the relevant dataset, and six of them contributed to improved performance metrics. Among these, the Sobel and Scharr methods achieved the highest test accuracy on unseen images (99.91%), while Scharr and Canny showed the lowest test loss values (0.29%). Furthermore, the Sobel and Scharr methods showed improvements in performance, achieving IoU score increases of 2.01% and 1.87%, respectively, over the baseline U-Net model without preprocessing, which is important for meeting the project's goal of improving the accuracy of facial skin temperature measurements. By integrating these edge detection techniques, the accuracy of facial temperature measurements was enhanced, enabling the precise identification of hot flush events. This research underscores the value of combining advanced image processing methods with deep learning models for enhancing the analysis of thermal images in biomedical applications.</p>