Deep learning algorithms to classify Fitzpatrick skin types for smartphone based NIRS imaging device
Conference
Leizaola, D, Sobhan, M, Kaile, K et al. (2023). Deep learning algorithms to classify Fitzpatrick skin types for smartphone based NIRS imaging device
. SMART BIOMEDICAL AND PHYSIOLOGICAL SENSOR TECHNOLOGY XI, 12516 10.1117/12.2665179
Leizaola, D, Sobhan, M, Kaile, K et al. (2023). Deep learning algorithms to classify Fitzpatrick skin types for smartphone based NIRS imaging device
. SMART BIOMEDICAL AND PHYSIOLOGICAL SENSOR TECHNOLOGY XI, 12516 10.1117/12.2665179
Non-contact imaging modalities for monitoring wound health could supplement the current standard which is a visual inspection by clinicians. Recently, a smartphone oxygenation tool (SPOT) has been developed for physiological imaging of tissue oxygenation changes in response to treatment. However, upon visual inspection of the wound bed and surrounding area, there are variations of pigmentation. Melanin concentration is a highly absorbing chromophore that can impact spatial oxygenation measurements. The objective of this study was to classify the six Fitzpatrick Skin Types (FST) by applying deep learning techniques prior to correcting SPOT’s oxygenation maps. In this IRB-approved study, control subjects were imaged on seven skin locations of varying FST (I-VI) under three different lighting conditions using SPOT device’s camera. A benchmark dataset with samples of 28 × 28 pixel images of human subjects’ feet was developed in three color spaces (R-G-B, Y-Cb-Cr, L-a-b). A deep learning algorithm, specifically a convolutional neural network (CNN), was used to classify skin into six FST (classes). Preliminary results showed the skin types on control subjects’ feet could be classified using deep learning with hyperparameter tuning with accuracies of > 82%. Our ongoing efforts are focused on extensive in-vivo studies on control subjects of FST I-VI on feet towards future implementation of the technology for diabetic wounds and oxygenation mapping using the SPOT device.
