Nudrat Nida1, Aun Irtaza2, Ali Javed3, Muhammad Haroon Yousaf4, Muhammad Tariq Mahmood5. 1. Department of Computer Engineering, University of Engineering & Technology, Taxila, Pakistan. Electronic address: 16F-PHD-CP-53@uettaxila.edu.pk. 2. Department of Computer Science, University of Engineering & Technology, Taxila, Pakistan. Electronic address: aun.irtaza@uettaxila.edu.pk. 3. Department of Software Engineering, University of Engineering & Technology, Taxila, Pakistan. Electronic address: ali.javed@uettaxila.edu.pk. 4. Department of Computer Engineering, University of Engineering & Technology, Taxila, Pakistan. Electronic address: haroon.yousaf@uettaxila.edu.pk. 5. School of Computer Science and Engineering, Korea University of Technology and Education, 1600, Chungjeol-ro, Byeongcheon-myeon, Dongnam-gu, Cheonan-si, Chungcheongnam-do, 31253, Republic of Korea. Electronic address: tariq@koreatech.ac.kr.
Abstract
OBJECTIVE: Melanoma is a dangerous form of the skin cancer responsible for thousands of deaths every year. Early detection of melanoma is possible through visual inspection of pigmented lesions over the skin, treated with simple excision of the cancerous cells. However, due to the limited availability of dermatologists, the visual inspection alone has the limited and variable accuracy that leads the patient to undergo a series of biopsies and complicates the treatment. In this work, a deep learning method is proposed for automated Melanoma region segmentation using dermoscopic images to overcome the challenges of automated Melanoma region segmentation within dermoscopic images. MATERIALS AND METHODS: A deep region based convolutional neural network (RCNN) precisely detects the multiple affected regions in the form of bounding boxes that simplify localization through Fuzzy C-mean (FCM) clustering. Our method constitutes of three step process: skin refinement, localization of Melanoma region, and finally segmentation of Melanoma. We applied the proposed method on benchmark dataset ISIC-2016 by International Symposium on biomedical images (ISBI) having 900 training and 376 testing Melanoma dermatological images. MAIN FINDINGS: The performance is evaluated for Melanoma segmentation using various quantitative measures. Our method achieved average values of pixel level specificity (SP) as 0.9417, pixel level sensitivity (SE) as 0.9781, F1 _ s core as 0.9589, pixel level accuracy (Ac) as 0.948. In addition, average dice score (Di) of segmentation was recorded as 0.94, which represents good segmentation performance. Moreover, Jaccard coefficient (Jc) averaged value on entire testing images was 0.93. Comparative analysis with the state of art methods and the results have demonstrated the superiority of the proposed method. CONCLUSION: In contrast with state of the art systems, the RCNN is capable to compute deep features with amen representation of Melanoma, and hence improves the segmentation performance. The RCNN can detect features for multiple skin diseases of the same patient as well as various diseases of different patients with efficient training mechanism. Series of experiments towards Melanoma detection and segmentation validates the effectiveness of our method.
OBJECTIVE:Melanoma is a dangerous form of the skin cancer responsible for thousands of deaths every year. Early detection of melanoma is possible through visual inspection of pigmented lesions over the skin, treated with simple excision of the cancerous cells. However, due to the limited availability of dermatologists, the visual inspection alone has the limited and variable accuracy that leads the patient to undergo a series of biopsies and complicates the treatment. In this work, a deep learning method is proposed for automated Melanoma region segmentation using dermoscopic images to overcome the challenges of automated Melanoma region segmentation within dermoscopic images. MATERIALS AND METHODS: A deep region based convolutional neural network (RCNN) precisely detects the multiple affected regions in the form of bounding boxes that simplify localization through Fuzzy C-mean (FCM) clustering. Our method constitutes of three step process: skin refinement, localization of Melanoma region, and finally segmentation of Melanoma. We applied the proposed method on benchmark dataset ISIC-2016 by International Symposium on biomedical images (ISBI) having 900 training and 376 testing Melanoma dermatological images. MAIN FINDINGS: The performance is evaluated for Melanoma segmentation using various quantitative measures. Our method achieved average values of pixel level specificity (SP) as 0.9417, pixel level sensitivity (SE) as 0.9781, F1 _ s core as 0.9589, pixel level accuracy (Ac) as 0.948. In addition, average dice score (Di) of segmentation was recorded as 0.94, which represents good segmentation performance. Moreover, Jaccard coefficient (Jc) averaged value on entire testing images was 0.93. Comparative analysis with the state of art methods and the results have demonstrated the superiority of the proposed method. CONCLUSION: In contrast with state of the art systems, the RCNN is capable to compute deep features with amen representation of Melanoma, and hence improves the segmentation performance. The RCNN can detect features for multiple skin diseases of the same patient as well as various diseases of different patients with efficient training mechanism. Series of experiments towards Melanoma detection and segmentation validates the effectiveness of our method.
Authors: Mehwish Dildar; Shumaila Akram; Muhammad Irfan; Hikmat Ullah Khan; Muhammad Ramzan; Abdur Rehman Mahmood; Soliman Ayed Alsaiari; Abdul Hakeem M Saeed; Mohammed Olaythah Alraddadi; Mater Hussen Mahnashi Journal: Int J Environ Res Public Health Date: 2021-05-20 Impact factor: 3.390