Two-stage Skin Lesion Segmentation from Dermoscopic Images by Using Deep Neural Networks
محل انتشار: فصلنامه زیست پزشکی جرجانی، دوره: 8، شماره: 2
سال انتشار: 1399
نوع سند: مقاله ژورنالی
زبان: انگلیسی
مشاهده: 283
فایل این مقاله در 15 صفحه با فرمت PDF قابل دریافت می باشد
- صدور گواهی نمایه سازی
- من نویسنده این مقاله هستم
این مقاله در بخشهای موضوعی زیر دسته بندی شده است:
استخراج به نرم افزارهای پژوهشی:
شناسه ملی سند علمی:
JR_JOBJ-8-2_008
تاریخ نمایه سازی: 27 بهمن 1399
چکیده مقاله:
Background and objective: Automatic semantic segmentation of skin lesions is one of the most important medical requirements in the diagnosis and treatment of skin cancer, and scientists always try to achieve more accurate lesion segmentation systems. Developing an accurate model for lesion segmentation helps in timely diagnosis and appropriate treatment.
Material and Methods: In this study, a two-stage deep learning-based method is presented for accurate segmentation of skin lesions. At the first stage, detection stage, an approximate location of the lesion in a dermoscopy is estimated using deep Yolo v2 network. A sub-image is cropped from the input dermoscopy by considering a margin around the estimated lesion bounding box and then resized to a predetermined normal size. DeepLab convolutional neural network is used at the second stage, segmentation stage, to extract the exact lesion area from the normalized image.
Results: A standard and well-known dataset of dermoscopic images, (ISBI) 2017 dataset, is used to evaluate the proposed method and compare it with the state-of-the-art methods. Our method achieved Jaccard value of 79.05%, which is 2.55% higher than the Jaccard of the winner of the ISIC 2017 challenge.
Conclusion: Experiments demonstrated that the proposed two-stage CNN-based lesion segmentation method outperformed other state-of-the-art methods on the well-known ISIB2017 dataset. High accuracy in detection stage is of most important. Using the detection stage based on Yolov2 before segmentation stage, DeepLab3+ structure with appropriate backbone network, data augmentation, and additional modes of input images are the main reasons of the significant improvement.
کلیدواژه ها:
نویسندگان
Fatemeh Bagheri
Department of Industrial Engineering, K. N. Toosi University of Technology, Pardis Street, Molla Sadra Ave, Tehran, Iran
Mohammad Jafar Tarokh
Department of Industrial Engineering, K. N. Toosi University of Technology, Pardis Street, Molla Sadra Ave, Tehran, Iran
Majid Ziaratban
Department of Electrical Engineering, Faculty of Engineering, Golestan University, Gorgan, Iran
مراجع و منابع این مقاله:
لیست زیر مراجع و منابع استفاده شده در این مقاله را نمایش می دهد. این مراجع به صورت کاملا ماشینی و بر اساس هوش مصنوعی استخراج شده اند و لذا ممکن است دارای اشکالاتی باشند که به مرور زمان دقت استخراج این محتوا افزایش می یابد. مراجعی که مقالات مربوط به آنها در سیویلیکا نمایه شده و پیدا شده اند، به خود مقاله لینک شده اند :