Thermal Image based Non Invasive Disease Diagnosis using Nature Inspired Algorithms

Passive infrared thermography offers a non-invasive, radiation-free modality for early detection of pathological conditions by capturing subtle surface-temperature anomalies. We propose a unified diagnostic framework that integrates automated ROI segmentation with validation, generative and geometric data augmentation to address class imbalance, deep-feature extraction, and metaheuristic feature selection to screen for breast cancer, diabetic foot ulcers, and thyroid nodules from thermal images. Generative data augmentation techniques systematically enlarge limited thermal imaging datasets to enhance model generalizability and reduce limitations in datasets. Techniques in ROI segmentation enable accurate detection of thermal differences between various disease categories. Deep learning neural models help in decoding intricate thermal patterns, which in turn transforms passive thermography from being an auxiliary diagnostic tool into a potential preliminary screening tool. On five public datasets - two of breast cancer, two of diabetic foot ulcers, and one of thyroid nodules; our method achieves 98.5%, 99.2%, and 97.85% accuracy, respectively, representing 3–5% absolute gains over unoptimized baselines. The present work brings together metaheuristic nature-inspired algorithms and computational intelligence to formulate a robust diagnostic platform adaptable for use in different domains of pathology, demonstrating considerable potential for non-invasive medical screening and real-time early detection strategies.