A deep learning approach to detect blood vessels in basal cell carcinoma

Abstract

Purpose: Blood vessels called telangiectasia are visible in skin lesions with the aid of dermoscopy. Telangiectasia are a pivotal identifying feature of basal cell carcinoma.These vessels appear thready, serpiginous, and may also appear arborizing, that is, wide vessels branch into successively thinner vessels. Due to these intricacies, their detection is not an easy task, neither with manual annotation nor with computerized techniques. In this study, we automate the segmentation of telangiectasia in dermoscopic images with a deep learning U-Net approach. Methods:We apply a combination of image processing techniques and a deep learning-based U-Net approach to detect telangiectasia in digital basal cell carcinoma skin cancer images. We compare loss functions and optimize the performance by using a combination loss function to manage class imbalance of skin versus vessel pixels. Results:We establish a baseline method for pixel-based telangiectasia detection in skin cancer lesion images. An analysis and comparison for human observer variability in annotation is also presented. Conclusion:Our approach yields Jaccard score within the variation of human observers as it addresses a new aspect of the rapidly evolving field of deep learning: automatic identification of cancer-specific structures. Further application of DL techniques to detect dermoscopic structures and handle noisy labels is warranted.

The application of artificial intelligence in the detection of basal cell carcinoma: A systematic review

“…Twelve articles reported on the detection of BCC in dermoscopy images 30‐41 . Half of the studies included used retrospectively collected datasets, such as ISIC‐2018 and ISIC‐2019, which contain multiple skin conditions including malignant lesions such as BCC and benign conditions 42,43 .…”

“…Two articles reported on the accuracy for multiple classes; hence, we calculated sensitivity and specificity from the provided confusion matrix (Table 1). 30,31 The accuracy (for binary outcome) was reported in one study, 37 while one study 39 reported the detection rate as the main outcome. Two studies 32,34 developed a binary classier (BCC vs no‐BCC) using relatively comparable data set sizes.…”

“…Twelve articles reported on the detection of BCC in dermoscopy images. [30][31][32][33][34][35][36][37][38][39][40][41] Half of the studies included used retrospectively collected datasets, such as ISIC-2018 and ISIC-2019, which contain multiple skin conditions including malignant lesions such as BCC and benign conditions. 42,43 The number of images used to train the model varied between studies, with only one publication using less than 1000 images to train and test the model.…”

“…Two studies 35,37 approached BCC detection in dermoscopy images by segmenting the vascular structures (telangiectasia). Kharazmi et al used an unsupervized learning algorithm (K-means clustering) to categorize unlabelled data depending on feature similarity.…”

Agreement Between Experts and an Untrained Crowd for Identifying Dermoscopic Features Using a Gamified App: Reader Feasibility Study

“…Machine learning (ML) methods have recently been investigated in the field of dermatology, and the majority of developed algorithms are diagnostic binary classifiers [11,12]. A number of studies have evaluated the performance of algorithms developed to detect specific dermoscopic features, including pigment network structures, vessels, and blue-white veil; however, many algorithms were trained and tested on relatively small data sets and have achieved only moderate accuracy [13][14][15][16][17][18][19][20][21].…”