Saturday 01 February 2025
Skin cancer is a growing concern worldwide, and early detection is crucial for effective treatment. One of the key challenges in diagnosing skin cancer is segmenting affected skin regions from normal tissue. This process typically involves analyzing high-resolution images using artificial intelligence-powered devices.
Researchers have developed various segmentation models to tackle this problem, but most are computationally expensive and require large amounts of data. A new model called MambaU-Lite has been designed to address these limitations while still delivering accurate results.
MambaU-Lite is a lightweight model that combines the strengths of two existing architectures: Mamba and convolutional neural networks (CNNs). The model uses a novel component called P-Mamba, which incorporates vision state space blocks alongside multiple pooling layers. This allows MambaU-Lite to effectively learn multi-scale features and enhance segmentation performance.
The researchers evaluated MambaU-Lite on two skin lesion datasets: ISIC 2018 and PH2. The results showed that the model outperformed other existing models, including U-Net, Attention U-Net, UNeXt, DCSAU-Net, and U-Lite. MambaU-Lite achieved a Dice similarity coefficient (DSC) of 0.9057 and an intersection over union (IoU) of 0.8361 on the ISIC 2018 dataset, while achieving a DSC of 0.9572 and an IoU of 0.9189 on the PH2 dataset.
One of the key advantages of MambaU-Lite is its ability to learn features at multiple scales. The P-Mamba block allows the model to capture both high-level and fine-grained details, making it well-suited for skin lesion segmentation. Additionally, the model’s lightweight architecture makes it suitable for deployment on medical devices.
The development of MambaU-Lite highlights the importance of balancing performance with computational efficiency in medical image analysis. As the healthcare industry continues to rely on artificial intelligence to improve diagnosis and treatment outcomes, researchers will need to prioritize models that can deliver accurate results while also being feasible for real-world implementation.
Cite this article: “MambaU-Lite: A Lightweight Model for Accurate Skin Lesion Segmentation”, The Science Archive, 2025.
Skin Cancer, Segmentation, Mambau-Lite, Artificial Intelligence, Convolutional Neural Networks, Cnns, P-Mamba, Vision State Space Blocks, Medical Image Analysis, Lightweight Model
