Color-invariant skin lesion semantic segmentation based on modified U-Net deep convolutional neural network.

Area of Science:

• Medical imaging
• Artificial intelligence
• Dermatology

Background:

• Melanoma is a dangerous, fast-growing skin cancer with increasing mortality rates.
• Accurate and timely diagnosis of melanoma from dermoscopy images is crucial but challenging.
• Current deep learning models often struggle with subtle lesion details using standard RGB color models.

Purpose of the Study:

• To develop an improved deep learning model for accurate skin lesion segmentation.
• To enhance the performance of diagnostic systems for melanoma detection.
• To overcome limitations of RGB color models in dermoscopy image analysis.

Main Methods:

• Proposed a novel Color Invariant U-Net (CIU-Net) architecture incorporating a color mixture block and a channel-attention module.
• Integrated various color models to create a new, informative three-channel representation.
• Utilized a hybrid loss function to optimize segmentation performance.

Main Results:

• The CIU-Net model demonstrated superior performance in segmenting skin lesions.
• Achieved state-of-the-art Dice and Jaccard coefficients of 92.56% and 91.40% on the ISIC 2018 dataset.
• Outperformed existing recent approaches in segmentation accuracy.

Conclusions:

• The CIU-Net model significantly enhances skin lesion segmentation accuracy.
• The proposed approach offers a promising tool for computer-aided melanoma diagnosis.
• Color invariant features and attention mechanisms are effective for improving deep learning in medical imaging.