Table 1 Comparative analysis of skin cancer detection research with segmentation and classification focus.
From: Advancing skin cancer diagnosis with deep learning and attention mechanisms
Reference | Key Method | Dataset(s) | Key Innovation | Methodological Gaps | Future Directions | Clinical Applicability | Model Efficiency | Computational Cost | Robustness | Real-time Performance |
|---|---|---|---|---|---|---|---|---|---|---|
Ozdemir & Pacal3 | ConvNeXtV2 with Self-attention | ISIC | Integration of ConvNeXtV2 with self-attention for improved melanoma detection | Limited to melanoma detection, lacks multi-class capability | Test on multi-class skin lesion datasets and enable real-time clinical usage | Primarily for melanoma detection | Moderate | Moderate | Moderate | High |
Nawaz et al.4 | CNN-based Deep Learning | ISIC | Dermoscopic image analysis for skin cancer detection using CNN models | Restricted to dermoscopic images, lacks rare conditions detection | Expand to include rare skin lesions and non-dermoscopic images | Limited to static analysis | Moderate | Moderate | High | Moderate |
Pacal et al.,6 | CNN-ViT Hybrid Model | ISIC | Hybrid approach combining CNNs and ViTs for early skin cancer diagnosis | Focus primarily on melanoma, limited to the ISIC dataset | Evaluate on multi-class skin lesions and extend to clinical settings | Real-world clinical use is feasible | High | High | High | High |
Kandhro et al.10 | E-VGG19 | ISIC | Enhanced VGG19 for real-time skin cancer detection | Trade-off between real-time performance and segmentation accuracy | Focus on minimizing latency for mobile and resource-constrained devices | Real-time application possible | High | High | High | High |
Li et al.16 | DSEUNet | ISIC | Lightweight UNet for dynamic space grouping enhancement for skin lesion segmentation | High computational cost for real-time use | Focus on reducing computational overhead and enabling mobile device usage | High efficiency for small devices | Low | Low | High | High |
Ahamed et al.17 | UNet with Attention Guidance | ISIC | Use of attention-guided UNet to enhance segmentation quality | Not evaluated on skin tone diversity, and limited real-world validation | Improve robustness for clinical applications and real-time deployment | Applicable to clinical use | Moderate | Moderate | Moderate | Moderate |
Wu et al.11 | MHorUNet | ISIC | High-order spatial interaction model for skin lesion segmentation | Limited to the ISIC dataset, lacks real-world deployment validation | Integration with multimodal datasets and real-time IoT-based sensor data | Limited to image-based analysis | Moderate | Moderate | Moderate | Moderate |
Wu et al.12 | HSH-UNet | ISIC | Hybrid high-order interactive model for enhanced segmentation performance | Does not generalize well to diverse datasets and real-time applications | Test on diverse datasets and deploy for real-time clinical use | Primarily for research use | High | High | High | Moderate |
Al-Waisy et al.39 | Deep Learning Framework | Dermoscopy | Early diagnosis and classification of skin cancer lesions | Only tested on dermoscopic, lacks real-time monitoring | Expand to real-time monitoring and multi-modal sensor integration | Suitable for research and clinical testing | Moderate | High | High | High |
