An improved skin lesion detection solution using multi-step preprocessing features and NASNet transfer learning model

• Published in: Image and vision computing Vol. 144; p. 104969
• Main Authors: Altamimi, Abdulaziz, Alrowais, Fadwa, Karamti, Hanen, Umer, Muhammad, Cascone, Lucia, Ashraf, Imran
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.04.2024
• Subjects: Computer vision application; Computer-aided diagnosis; Dermatology pigmented lesion classification; Multi-step image processing; NASNet model; Skin lesion; Transfer learning; Skin lesion; NASNet model; Transfer learning; Computer vision application; Multi-step image processing; Dermatology pigmented lesion classification; Computer-aided diagnosis
• ISSN: 0262-8856; 1872-8138
• DOI: 10.1016/j.imavis.2024.104969

Computer-aided diagnosis has shown its potential for accurate detection of various diseases like skin lesion. Skin lesion has been recognized as a challenging task since manual identification through visual analysis of images can be inefficient, tedious, and error-prone. Although automatic diagnosis approaches are used to overcome this challenge, it is crucial to address problems such as variations in the size of images, presence of hairs in images, unsatisfactory schemes of colors, ruler markers, low-contrast, and differences in dimensions of lesions, and gel bubbles. Researchers in the field of dermatology pigmented lesion classification have proposed different methodologies to confront this issue. Specifically, they have focused on the binary classification problem of distinguishing Melanocytic lesions from normal ones. In this research, the dataset “MNIST HAM10000” is utilized, published by International Skin Image Collaboration, and contains data about 07 different skin cancer types. Moreover, in this research, we have focused on image preprocessing and skin lesion detection with NASNet model. Experimental reuslts demonstrated the superiority of the proposed model, which achieves an accuracy of 99.85%. This accomplishment has been made possible with the utilization of data augmentation techniques and multi-step image processing methods with the proposed NasNET model.