Dermoscopic image segmentation method based on convolutional neural networks

Dang Ngoc Hoang Thanh D. N. H. T., Le Thi Thanh L. T. T., ERKAN U., Khamparia A., Prasath V. B. S.

INTERNATIONAL JOURNAL OF COMPUTER APPLICATIONS IN TECHNOLOGY, cilt.66, sa.2, ss.89-99, 2021 (ESCI, Scopus) identifier identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 66 Sayı: 2
  • Basım Tarihi: 2021
  • Doi Numarası: 10.1504/ijcat.2021.119757
  • Dergi Adı: INTERNATIONAL JOURNAL OF COMPUTER APPLICATIONS IN TECHNOLOGY
  • Derginin Tarandığı İndeksler: Emerging Sources Citation Index (ESCI), Scopus, PASCAL, ABI/INFORM, Aerospace Database, Applied Science & Technology Source, Business Source Elite, Business Source Premier, Communication Abstracts, Computer & Applied Sciences, INSPEC, Metadex, Civil Engineering Abstracts
  • Sayfa Sayıları: ss.89-99
  • Ankara Üniversitesi Adresli: Hayır

Özet

In this paper, we present an efficient dermoscopic image segmentation method based on the linearisation of gamma-correction, and convolutional neural networks. Linearisation of gamma-correction is helpful to enhance low-intensity regions of skin lesion areas. Therefore, postprocessing tasks can work more effectively. The proposed convolutional neural network architecture for the segmentation method is based on the VGG-19 network. The acquired training results are convenient to apply the semantic segmentation method. Experimental results are conducted on the public ISIC-2017 dataset. To assess the quality of obtained segmentations, we make use of standard error metrics such as the Jaccard and Dice which are based on the overlap with ground truth, along with other measures such as the accuracy, sensitivity, and specificity. Moreover, we provide a comparison of our segmentation results with other similar methods. From experimental results, we infer that our method obtains excellent results in all the metrics and obtains competitive performance over other current and state of the art models for dermoscopic image segmentation.