Skin cancer classification using YOLO-NAS based convolutional network with modified focal loss

---

Abstract

This work presents a novel approach for skin lesion classification through the utilization of a YOLONAS (You Only Look Once Neural Architecture Search) based convolutional network. Leveraging the speed and efficiency of YOLO's object detection algorithm and the automated architecture design capabilities of NAS, the system achieves advanced levels of classification accuracy while maintaining real-time processing capabilities. By seamlessly integrating object detection and neural architecture search, the proposed YOLO-NAS based convolutional network with modified focal loss offers a promising solution for efficient and effective skin lesion classification, thereby contributing to improved early diagnosis and timely medical interventions for dermatological conditions. On the ISIC-2019 dataset, this technique obtains an accuracy of 92.00%, an average precision of 92.12%, an average recall of 92.37%, and an average F1-score of 92.12%.

---

Keywords

Deep Learning, Skin lesion classification, HAM10000 dataset, Skin lesion classification