An optimized CNN based intelligent prognostics model for disease prediction and classification from Dermoscopy images

• Published in: Multimedia tools and applications Vol. 79; no. 35-36; pp. 26817 - 26835
• Main Authors: Tyagi, Ankita, Mehra, Ritika
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.09.2020; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Artificial neural networks; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Datasets; Diagnostic systems; Image analysis; Image classification; Image segmentation; Lesions; Medical imaging; Multimedia Information Systems; Particle swarm optimization; Skin cancer; Special Purpose and Application-Based Systems; Vessels; Speed up robust features (SURF) descriptor; Skin lesion segmentation; Lesion hair removal; Convolutional neural network (CNN); Particle swarm optimization (PSO); Intelligent prognostics model
• ISSN: 1380-7501; 1573-7721
• DOI: 10.1007/s11042-020-09074-3

Tiny skin vessels and telangiectasia are most imperative dermoscopy configurations used to differentiate Basal Cell Carcinoma (BCC) from benign skin lesions. This research work builds off of previously developed image analysis techniques to identify vessels automatically to separate benign lesions from BCCs. In this paper, to develop a model for Intelligent Prognostics Model for Disease Prediction and Classification (IPM-DPC) from dermoscopy images is presented using the combination of Convolutional Neural Network (CNN) structure along with the Particle Swarm Optimization (PSO). Here PSO play two different roles in this proposed IPM-DPC, firstly PSO used with K-means segmentation technique to improve the segmentation accuracy then PSO is used as filter for the CNN to train the proposed IPM-DPC. Speed up Robust Features (SURF) algorithm is used as feature descriptor along with PSO as feature selection algorithm which increase the classification accuracy of the system. This study uses a dataset of 1000 dermoscopy skin lesion images of 545 BCCs and 455Non-BCCs or benign images as the input sets. This dataset is taken from ISBI-2016 Dataset and available on: www.isic-archive.com . Experimental results yielded a diagnostic accuracy as high as 99.46% using the IMP-DPC approach, providing a14.94% improvement over a system without using the PSO as filter layer in CNN. When the evaluation parameters of proposed IMP-DPC is compared with a few other state-of-art methods, the proposed method achieves the best performance in terms of accuracy and detection time in differentiating BCC and Non-BCC from dermoscopy skin lesion images .