•  
  •  
 

Corresponding Author

Mohammed M. Abo-Zahhad

Subject Area

Biomedical Engineering

Article Type

Original Study

Abstract

Detecting skin cancer early and accurately is crucial for successfully treating this potentially fatal disease. Enhancing the accuracy of visual inspection techniques is often necessary to improve clinical decision-making and increase the chance of successful treatment outcomes. In this research, high-performance deep learning (DL) models for automated skin cancer categorization and early skin cancer diagnosis screening are developed and evaluated in conjunction with the discrete cosine transform (DCT). For this purpose, features are extracted from medical images using both techniques. The DCT is used for feature extraction and dimensionality reduction, while deep learning (DL) trains fully connected models for classification and feature combination. The proposed efficient skin cancer detection algorithm integrates features derived from dermoscopic images through DCT and deep learning models to yield rapid detection with better performance measures. As a result, dermoscopy images facilitate automated early detection instead of the labor-intensive, time-consuming method that demands considerable highly skilled expertise.

Pre-trained DCT-VGG-16 and DCT-ResNet50 models using a sizable dataset of dermoscopy images from ISIC 2019 are utilized for high-accuracy classification at minimal computational cost. Twenty-five thousand three hundred thirty-one dermoscopy images in eight different diagnostic categories are adopted. The training loss, training accuracy, validation loss, validation accuracy, sensitivity, and specificity are performance measures for reliable model training and validation. Results indicate that DCT-VGG16 performs better than DCT-ResNet-50 in terms of sensitivity and accuracy; however, DCT-ResNet-50 requires a more extended training period than DCT-VGG16. However, DCT-ResNet-50 yields a lesser accuracy of 96.5% and sensitivity of 95.1%, DCT-VGG16 attained a higher accuracy of 99.4% and sensitivity of 97.6%. Comparison of DCT-ResNet50 and DCT-VGG16 on the HAM10000 dataset reveals the superior performance of DCT-VGG16 over ten state-of-the-art recently published methods and comparable accuracy of DCT-ResNet50 to the most recent one. So, the proposed methods help dermatologists and medical professionals to automate disease identification and improve patient care.

Keywords

Melanoma, Deep Learning, CNN, DCT, ISIC Datasets, Skin Cancer Diagnosis

Creative Commons License

Creative Commons Attribution 4.0 License
This work is licensed under a Creative Commons Attribution 4.0 License.

Download

Share

COinS