Automated Skin Lesion Classification on Ultrasound Images

Abstract

The growing incidence of skin cancer makes computer-aided diagnosis tools for this group of diseases increasingly important. The use of ultrasound has the potential to complement information from optical dermoscopy. The current work presents a fully automatic classification framework utilizing fully-automated (FA) segmentation and compares it with classification using two semi-automated (SA) segmentation methods. Ultrasound recordings were taken from a total of 310 lesions (70 melanoma, 130 basal cell carcinoma and 110 benign nevi). A support vector machine (SVM) model was trained on 62 features, with ten-fold cross-validation. Six classification tasks were considered, namely all the possible permutations of one class versus one or two remaining classes. The receiver operating characteristic (ROC) area under the curve (AUC) as well as the accuracy (ACC) were measured. The best classification was obtained for the classification of nevi from cancerous lesions (melanoma, basal cell carcinoma), with AUCs of over 90% and ACCs of over 85% obtained with all segmentation methods. Previous works have either not implemented FA ultrasound-based skin cancer classification (making diagnosis more lengthy and operator-dependent), or are unclear in their classification results. Furthermore, the current work is the first to assess the effect of implementing FA instead of SA classification, with FA classification never degrading performance (in terms of AUC or ACC) by more than 5%.

Computer-Aided Diagnosis Methods for High-Frequency Ultrasound Data Analysis: A Review

“…From the segmented regions and the surrounding tissues, the authors extracted 93 features (textural- and shape-based) from which they finally selected 62 for the SVM-based classification step. Compared to the similar work by Csabai et al [ 82 ], the authors obtained [ 21 ] better classification results. It is worth mentioning that the authors shared all the code used in this work on GitHub [ 91 ].…”

“…The second most widely explored area was skin tumor segmentation [ 20 , 26 , 63 , 75 , 76 , 78 ]. In support of skin tumor diagnosis, there also appeared works targeting HFUS image classification [ 21 , 45 , 46 , 81 , 82 , 83 ], with a particular emphasis on melanocytic lesion classification. However, other classification problems were also found in the literature [ 74 , 86 ].…”

“…The newest framework developed by Marosán-Vilimszky et al [ 21 ] utilized the benefits of the seeding selection step and segmentation algorithm described in [ 20 ] for the skin lesion classification problem and compared it with the semi-automated approach.…”

“…The newest work by Marosán-Vilimszky et al [ 21 ] was the last to focus on nevus. The analyzed HFUS recordings included 310 lesions comprising 70 melanomas, 130 basal cell carcinomas, and 110 benign nevi.…”

“…There are two available transducers: 20 MHz, with a 60 by 150-micron resolution and in-depth penetration of 15 mm or a 60 by 260-micron resolution and a 23 mm penetration, as well as 50 MHz providing a 30 by 60-micron resolution and a 3 mm max penetration. The third manufacturer referred to in the literature [ 20 , 21 ] is Hitachi (Tokyo, Japan). The high-frequency machine HI VISION Preirus is equipped with a 5–18 MHz transducer (EUP-L75).…”