An IoMT-Based Melanoma Lesion Segmentation Using Conditional Generative Adversarial Networks.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Internet of Medical Things (IOMT) enables remote data collection and medical assistance.
• AI and Deep Learning in IOMT devices facilitate Computer-Aided Diagnosis (CAD) systems for diseases like melanoma.
• Accurate segmentation of skin lesions is crucial for effective melanoma diagnosis, but visual similarity poses a challenge.

Purpose of the Study:

• To propose an advanced generative deep learning model for accurate segmentation of melanoma skin lesions.
• To address the limitations of existing methods in segmenting visually similar lesions.
• To enhance the accuracy of CAD systems for melanoma detection.

Main Methods:

• Development of a Conditional Generative Adversarial Network (cGAN) for lesion segmentation.
• The cGAN model generates segmented images conditioned on dermoscopic images of skin lesions.
• Validation of the proposed model using three distinct datasets: DermQuest, DermIS, and ISCI2016.

Main Results:

• The proposed cGAN model achieved optimal segmentation results across all tested datasets.
• Performance accuracy reached 99% on the DermQuest dataset.
• Accuracy of 97% and 95% was attained on the DermIS and ISCI2016 datasets, respectively.

Conclusions:

• The Conditional Generative Adversarial Network (cGAN) is an effective solution for accurate melanoma skin lesion segmentation.
• The proposed model demonstrates superior performance compared to traditional and other deep learning methods.
• This advancement holds significant potential for improving early melanoma diagnosis through AI-powered medical imaging.