Ridge and furrow pattern classification for acral lentiginous melanoma using dermoscopic images. (February 2017)

• Record Type: Journal Article
• Authors: Yang, Sejung; Oh, Byungho; Hahm, Sungwon; Chung, Kee-Yang; Lee, Byung-Uk
• Is Part Of: Biomedical signal processing and control. Volume 32(2017)
• Page Start: 90
• Page End: 96
• Publication Date: 2017-02
• Subjects: Acral lentiginous melanoma -- Image analysis -- Pattern classification -- Dermoscopic images; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2016.09.019
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: An automatic algorithm that can classify the "furrow" from "ridge" patterns of pigmentation on acral areas is proposed. Accurate discrimination is performed using the histogram analysis of dark/bright pattern width ratio. Boundary of a furrow/ridge patterns are reliably detected from local maxima of a Gaussian smoothed derivative of dermoscopy. Abstract: Background/purpose: The development of an automatic diagnostic algorithm using characteristics of dermoscopic findings in acral lentiginous melanoma (ALM) has been slow due to the rarity of melanoma in non-Caucasian populations. In this study, we present an automatic algorithm that can distinguish the "furrow" and "ridge" patterns of pigmentation on the palm and foot, and report its usefulness for the detection of ALM. Methods: To distinguish between ALM and nevus, the proposed image analysis is applied. From a dermoscopic image, edges having the steepest ascent or descent are detected through Gaussian derivative filtering. The widths between edges are then measured and the brightness of each stripe is tagged. The dark area is tagged as black and the bright area is tagged as white. The ratio of widths of dark to bright is calculated at each stripe pair and the histogram of the width ratio in the dermoscopic image is generated. Results: A total of 297 dermoscopic images confirmed by histopathologic diagnoses are classified. All of the melanoma dermoscopic images were classified correctly using the proposed algorithm, while only one nevus image was misclassified. The proposed method achieved a sensitivity of 100%, a specificity of 99.1%, an accuracy of 99.7%, and a similarity of 99.7%. Conclusion: In this study, we propose a novel automatic algorithm that can precisely distinguish the "furrow" and "ridge" patterns of pigmentation on dermoscopic images using the width ratio of dark and bright patterns. It is expected that the proposed algorithm will contribute to the early diagnosis of ALM.