Development of a Portable Optical Imaging System based on a Smartphone and Image Classification using a Learning Algorithm

Abstract

To date, various skin diseases have incrementally increases due to hereditary and environmental factors including the stress, irregular diet, pollution, and etc. Among diseases, seborrheic dermatitis and psoriasis are a chronic/relapsing dermatitis, which involve infection, temporary alopecia, and etc. To prevent complications and take appropriate prescription due to the diseases, it would be crucial to differentiate seborrheic dermatitis from psoriasis with high specificity and accuracy at the early stages as well as it would further necessary to continuously/quantitatively monitor the lesions during its treatment at locations besides a hospital. However, the discrimination between the diseases at the early stages would be challenging. Optical imaging techniques have been shown to have a crucial role to detect various skin diseases. Among them, an advanced dermoscope based on multispectral imaging techniques offers better specificity and sensitivity in the detection of skin lesions than a conventional RGB dermoscope. However, the advanced dermoscope utilized in the hospital is typically bulk and expensive and thus may not be suited for ubiquitous diagnosis and monitoring of skin lesions including seborrheic dermatitis and psoriasis. In this thesis, we here demonstrate a portable mobile multispectral imaging system attached to a smartphone with an external C-MOS camera and the potential learning-based classification method for detection of seborrheic dermatitis and psoriasis by using it. The system allowed to obtain images of skin lesions at nine consecutive wavelengths within the range of 400-700nm. It was here employed to perform multispectral imaging and analysis of lesions to discriminate between seborrheic dermatitis and psoriasis or other diseased regions. Also, the results classified by a RGB image classification, a spectral angle measure (SAM), and a multiclass classification method based on a learning algorithm were compared. It was here found that spectral signatures of seborrheic dermatitis and psoriasis were slightly different but they could be clearly discernable by their spectral signatures. The SAM and multiclass classification method offered better accuracy in discrimination between of seborrheic dermatitis and psoriasis occurring on the scalp than the RGB image classification method. These results suggested that the multispectral imaging and learning-based analysis may have the potential to discriminate between seborrheic dermatitis and psoriasis regions with high portability and acceptable specificity for mobile skin diagnosis. ⓒ 2017 DGIST