BLSNet: Skin lesion detection and classification using broad learning system with incremental learning algorithm. Issue 9 (12th February 2022)

• Record Type: Journal Article
• Authors: Gottumukkala, V. S. S. P. Raju; Kumaran, N.; Sekhar, V. Chandra
• Other Names: Herrero Álvaro guestEditor.; Urda Daniel guestEditor.; Sedano Javier guestEditor.; Quintián Héctor guestEditor.; Corchado Emilio guestEditor.; Ahmed Syed Hassan guestEditor.; Khan Murad guestEditor.; Guibene Wael guestEditor.
• Is Part Of: Expert systems. Volume 39:Issue 9(2022)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2022-02-12
• Subjects: broad learning system -- deep learning convolutional neural networks -- incremental learning algorithm -- melanoma and non‐melanoma classification -- skin lesion detection; Expert systems (Computer science); 006.33
• DOI: 10.1111/exsy.12938
• Languages: English
• ISSNs: 0266-4720

Abstract

Abstract: Background: Skin lesion detection and classification (SLDC) is extremely important in the diagnosis of skin cancer and detection of melanoma cancer. As a result, the use of image processing equipment integrated with artificial intelligence can assist dermatologists in their decision‐making and examination. In addition, all deep learning (DL) structures consumes more time due to the large number of associated factors in filters and layers. Furthermore, if the architecture is insufficient to prototype the classification system, it must go through a lengthy retraining procedure. Material and method: Therefore, this article proposes a broad learning system (BLS) using incremental learning algorithm for the classification of non‐melanoma and melanoma skin lesions from dermoscopic images. Here after the proposed model is termed as BLSNet. Results: Experiments on ISIC 2019 and PH 2 dataset indicate that proposed SLDC using BLSNet out‐perform the existing DL‐based SLDC models with an accuracy of 99.09% and F1 ‐score of 98.73%. Further, the overall execution time of proposed BLSNet is 0.93 s, which is superior as compared to the conventional approaches. Conclusion: Thus, the performance trade‐off between classification accuracy and execution time is achieved using proposed BLSNet model.