A novel tool for detecting Buruli ulcer disease based on multispectral image analysis on handheld devices

In this paper, we propose a new approach for detecting Buruli ulcer disease automatically based on computerized analysis of multispectral dermoscopic images. We employ optical skin modeling to reconstruct distribution maps of physiological tissue parameters that correlate well with melanin distribution in the epidermis as well as blood volume and blood oxygenation in the dermis. Using 197 sets of multispectral images-taken under while, blue, yellow, and red light-representing 16 Buruli and 181 non-Buruli lesions, we achieved overall classification sensitivity of 90.0%, specificity of 93.3%, and receiver operator characteristic area under the curve of 97.1%. These results suggest that automated multispectral image analysis implemented on a handheld device, such as the iPhone equipped with the appropriate attachment, can provide a fast, reliable, and economical tool to screen Buruli ulcer disease that can be easily deployed in countries where the disease is endemic and can help save lives.