The Color-Polarization Filter Array (CPFA) Sensors to Recognize Skin Lesions

The early diagnosis of skin diseases, specifically skin cancers, might have a crucial impact on people’s health, providing information enabling the immediate implementation of the appropriate therapy and eventually saving the patient’s life. The automation of the diagnosis process based on image analysis is broadly exploited in current research. Among others, the methods of polarization analysis in imaging systems have been growing in popularity in recent years. The image processing supported by the light polarisation-sensitive device has been used in numerous imaging applications, including medical imaging, machine vision, and autonomous vehicle navigation, to name a few. The main goal of this work is to develop a new solution for a comprehensive automated skin analysis system allowing the classification of skin lesions based on multimodal image data and deep machine learning models. The main challenge of this research is to improve the sensitivity and specificity of the melanoma diagnosis and finally determine the optimal configuration of the system for the acquisition of diagnostic data. In particular, a specific setup for polarisation image analysis will be presented and discussed in the context of its sensitivity and applicability. Additionally, the method of digital analysis and classification of acquired images will be presented and discussed in detail. As part of this work, the relevance of deep learning to enhance the recognition of human skin lesions was also investigated, with a specific focus on optimizing the method of using available image datasets for the initial training of neural network models.