article · Diagnostics
Existing skin cancer detection systems relying on pre-trained deep learning models often struggle to achieve high levels of diagnostic accuracy. To address this limitation, a framework has been developed that improves accuracy by extracting relevant image features using the MobileNetV3 architecture. These extracted features are subsequently refined using a modified Hunger Games Search algorithm that incorporates Particle Swarm Optimisation and Dynamic-Opposite Learning, designated as DOLHGS. This hybrid technique acts as a feature selection mechanism to isolate the most critical data points for classification. Evaluated on standard benchmark collections, the approach attained an accuracy of 88.19 percent on the two-class ISIC-2016 dataset and 96.43 percent on the three-class PH2 dataset. Experimental outcomes indicate that the methodology outperforms several widely used algorithms in both classification accuracy and feature optimisation.
Accurate identification of skin cancer from clinical imagery is essential for timely medical intervention. Standard automated detection systems often fail to isolate the most informative visual features, limiting their diagnostic precision. By pairing efficient deep learning models with advanced mathematical feature selection, this work demonstrates how automated screening tools can be made more accurate and computationally efficient when assessing skin lesions.
This framework could support diagnostic software tools designed for clinicians and dermatologists assessing suspected skin lesions. Because it relies on benchmark datasets ISIC-2016 and PH2, the technology represents early-stage, applied computational research. Substantial clinical validation on real-world patient data and integration into medical imaging devices or healthcare platforms would be needed before commercial deployment.
AI-generated from the published abstract. Always read the original work before citing.
Recently, pre-trained deep learning (DL) models have been employed to tackle and enhance the performance on many tasks such as skin cancer detection instead of training models from scratch. However, the existing systems are unable to attain substantial levels of accuracy. Therefore, we propose, in this paper, a robust skin cancer detection framework for to improve the accuracy by extracting and learning relevant image representations using a MobileNetV3 architecture. Thereafter, the extracted features are used as input to a modified Hunger Games Search (HGS) based on Particle Swarm Optimization (PSO) and Dynamic-Opposite Learning (DOLHGS). This modification is used as a novel feature selection to alloacte the most relevant feature to maximize the model's performance. For evaluation of the efficiency of the developed DOLHGS, the ISIC-2016 dataset and the PH2 dataset were employed, including two and three categories, respectively. The proposed model has accuracy 88.19% on the ISIC-2016 dataset and 96.43% on PH2. Based on the experimental results, the proposed approach showed more accurate and efficient performance in skin cancer detection than other well-known and popular algorithms in terms of classification accuracy and optimized features.
This page summarises published work. The authoritative version sits with the publisher.
DOI: 10.3390/diagnostics13091579
Is something wrong with this record? Report it or request removal.
Discussion
Have you built on this work, tried to replicate it, or seen it applied in practice? Share what you know. Verified researchers and MARATTO™ domain experts can open a discussion, and any member can reply. Contributions are reviewed before they appear.
No discussion yet. Open the first thread.
New to MARATTO™? Create a free account.
