InSiNet: a deep convolutional approach to skin cancer detection and segmentation

Area of Science:

• Dermatology and Artificial Intelligence
• Medical Imaging Analysis
• Computational Pathology

Background:

• Skin cancer is a significant global health concern, necessitating accurate and early diagnosis.
• Traditional skin cancer diagnosis relies heavily on expert interpretation, which can be subjective.
• Deep learning offers a promising approach to enhance the accuracy and objectivity of skin lesion classification.

Purpose of the Study:

• To develop and evaluate InSiNet, a deep learning-based convolutional neural network for detecting benign and malignant skin lesions.
• To compare the performance of InSiNet against established machine learning techniques.
• To assess the diagnostic accuracy of InSiNet on multiple benchmark datasets.

Main Methods:

• Development of InSiNet, a novel deep learning convolutional neural network architecture.
• Performance evaluation using the International Skin Imaging Collaboration (ISIC) HAM10000, ISIC 2019, and ISIC 2020 datasets.
• Comparative analysis of InSiNet against GoogleNet, DenseNet-201, ResNet152V2, EfficientNetB0, RBF-support vector machine, logistic regression, and random forest.

Main Results:

• InSiNet achieved high accuracy rates: 94.59% on ISIC 2018, 91.89% on ISIC 2019, and 90.54% on ISIC 2020.
• The developed InSiNet architecture demonstrated superior performance compared to other evaluated machine learning methods.
• The study confirmed the effectiveness of deep learning in improving the reliability of skin lesion diagnosis.

Conclusions:

• InSiNet represents a significant advancement in automated skin lesion classification.
• Deep learning models like InSiNet can augment traditional diagnostic methods, reducing human error.
• The findings support the integration of AI in dermatological practice for more accurate skin cancer detection.