A Comparative Study of HSV-Based and Machine Learning Methods for Automated Detection of Throat Inflammation from Images

This study explores and compares three image segmentation techniques—Static Thresholding, Adaptive Thresholding (Otsu and Gaussian), and K-Means Clustering— for classifying healthy and unhealthy throat images. Each method was evaluated based on its ability to segment pathological regions and extract meaningful visual features such as redness, whiteness, hue, saturation, and brightness. Statistical analysis revealed significant differences in key features between healthy and unhealthy samples, particularly in redness and whiteness ratios. Machine learning models, including Logistic Regression and Support Vector Machine (SVM), were applied to the extracted features to assess classification performance. Results show that Adaptive Thresholding, especially when paired with SVM, achieved the highest classification performance (ROC AUC = 0.62), indicating its effectiveness in distinguishing infection-related features. In contrast, K-Means Clustering performed poorly (ROC AUC < 0.5), suggesting limitations in clinical application for this segmentation method. These findings highlight the potential of adaptive techniques in automated throat infection detection and support their integration into clinical decision-support systems.