The introduction of artificial intelligence and neural networks within the medical industry has opened many opportunities for the improvement of the effectiveness, precision, and automation of previously tediously manual diagnostics processes. Nowadays, pre-trained artificial neural networks are widely used for various applications in the biomedical industry and are at the forefront of the digitalization of healthcare. In the case of skin lesion image processing, segmentation is required to align the multispectral images, to remove hair artifacts, as well as to perform automated parameter calculations within the skin formation for a multitude of skin lesion data categories. During the conducted research, artificial neural networks were used to segment skin lesions for automatic multispectral data processing. The network uses skin diffuse reflection images from 3 different spectral wavelengths. By using image processing methods, the input images were then segmented by the trained artificial neural network. The implementation of neural networks within the processing of multispectral skin cancer imagery solves most of the issues of previously conducted research and scripted solutions. The segmentation results have been improved in specific cases that cannot be processed using the previously developed script, such as images containing multiple markers, blurry, malformation areas with indistinguishable contours, as well as low contrast, misaligned or reflective imagery. To provide more training data for the artificial neural network and increase the segmentation process accuracy, the training image database has been increased by using data augmentation. Additional automation of multiple input image processing makes further improvement implementation and testing more accessible and interchangeable without the constraints of needing licensed software, whilst also making it possible to automate the processing of a large number of data to give an accurate assessment of skin lesions in the process of skin cancer diagnostics.
- 3.4. Other publications in conference proceedings (including local)