Automatic pigmented lesion segmentation through a dermoscopy-guided OCT approach for early diagnosis

Abstract

Early diagnosis of pigmented lesions, specially melanoma, is an unmet clinical need that would help to improve patient prognosis. Apart from histopathological biopsy, the only gold standard non-invasive imaging technique during diagnosis is dermatoscopy (DD). Over the last years, new medical imaging techniques are being developed and Optical Coherence Tomography (OCT) has demonstrated to be very helpful on dermatology. OCT is non-invasive and provides in-depth structural microscopic information of the skin in real-time. In comparison with other novel techniques, as Reflectance Confocal Microscopy (RCM), the acquisition time is lower and the field-of-view higher. Hence, consolidated diagnosis techniques and novel imaging modalities can be combined to improve decision making during diagnosis and treatment. With actual methods, the delineation of lesion margins directly on OCT images during early stages of the disease is still really challenging and, at the same time, relevant from a prognosis perspective. This work proposes combining DD and OCT images to take advantage of their complementary information. The goal is to guide lesions delineation on OCT images considering the clinical features on DD images. The developed method applies image processing techniques to DD image to automatically segment the lesion; later, and after a calibration procedure, DD and OCT images become coregistered. In a final step the DD segmentation is transferred into the OCT image. Applying advanced image processing techniques and the proposed strategy of lesion delimitation, histopathological characteristics of the segmented lesion can be studied on OCT images afterwards. This proposal can lead to early, real-time and non-invasive diagnosis of pigmented lesions.