Abstract:
Cellulite is a common and disturbing physiological condition of skin experienced by 85% to 98% of the post-pubertal females in developed countries and is classified as one of the worst tolerated by women deteriorating the quality of life. Despite the widespread information about cellulite and its pathogenesis, there is still a shortage of fast and widely available tools for its objective and early diagnosis. Although, the exact reason for cellulite is unknown, but it mostly depends on the physiology of subcutaneous adipose tissue or body fat. This subcutaneous adipose tissue creates edema and pressure which hampered the regular blood flow and create a temperature difference along the region of cellulite. On basis of this temperature difference, infrared thermographic image processing is used to propose a new approach for the identification of different stages of the cellulite which is fast, non-contact, non-invasive and can detect very early changes. We experimented on total 212 female volunteers, aged 19–22 with different Stages of cellulite and collected thermographic images of posterior site of thighs using a thermographic camera (FLIR T335). We propose an algorithm to pre-process the images and select the region of interest (ROI) automatically. As the volume of our dataset is low, we try to find out thebest combination of different feature extraction methods with different classification algorithms byexperimenting on the infrared thermographic images of cellulite. Combination of HOG (Histogram of oriented gradients) as the feature extraction methodand ANN (Artificial neural network) as the classification algorithm provides more than 80% average accuracy of all stages and for primary stage of cellulite it gives more than 90% sensitivity with an AUC (Area Under the Curve) of 0.8102. Our proposed system is fully automatic in the recognition of cellulite severity which will be able to work as preventive tooland provide fully objective personalized and cost-effective diagnosis.
Keywords: Infrared thermography, classification, feature extraction, artificial intelligence, cellulite, personal diagnosis
