Can deep learning accurately detect melanoma with limited computing power? This paper introduces DTONet (double-tailed octave network), a lightweight deep learning network specifically designed for melanoma segmentation. It is able to achieve high segmentation accuracy with minimal resource consumption. The model addresses the challenge of deploying effective diagnostic tools in resource-constrained environments, such as hospitals with limited hardware. With just 30,859 computational parameters, DTONet achieves superior performance compared to other models, with significant improvement in IOU, demonstrating good accuracy. To validate the generalization capability of this model, we conducted tests on the PH2 dataset, and the results still outperformed existing models. Its parameter count is only 1/256th of the mainstream UNet model. Therefore, the proposed DTONet network exhibits excellent generalization ability and is sufficiently outstanding. DTONet offers a promising solution for accurate and efficient melanoma segmentation, particularly in settings where computational resources are limited.
This paper aligns with Bioengineering's focus on engineering principles in biological systems. Presenting a lightweight deep learning network for melanoma segmentation, the study fits within the journal's scope, addressing the need for efficient diagnostic tools in healthcare using bioengineering approaches. The study is relevant to biomedical engineers seeking to develop practical solutions for medical imaging.