Related Experiment Video
Updated: May 24, 2025
Combining Reflectance Confocal Microscopy with Optical Coherence Tomography for Noninvasive Diagnosis of Skin Cancers via Image Acquisition
Published on: August 18, 2022
[An efficient and lightweight skin pathology detection method based on multi-scale feature fusion using an improved
Yuying Ren1, Lingxiao Huang1, Fang DU1
1School of Information Engineering, Ningxia University// Ningxia Key Laboratory of Artificial Intelligence and Information Security for Channeling Computing Resources from the East to the West//Collaborative Innovation Center for Ningxia Big Data and Artificial Intelligence Co-founded by Ningxia Municipality and Ministry of Education, Yinchuan 750021, China.
This study introduces an improved RT-DETR model for efficient and accurate skin disease detection. The enhanced model achieves higher accuracy and faster detection speeds while significantly reducing computational load and parameters.
Area of Science:
- Dermatology
- Computer Vision
- Artificial Intelligence
Context:
- Skin disease detection is challenged by multi-scale lesions, image noise, and limited diagnostic equipment resources.
- Existing models often struggle with efficiency and accuracy due to these complex factors.
- Accurate and rapid diagnosis is crucial for effective dermatological treatment.
Purpose:
- To develop a highly efficient and lightweight skin disease detection model.
- To improve the accuracy and speed of skin disease identification.
- To address the limitations of existing diagnostic tools through advanced deep learning techniques.
Summary:
- An improved RT-DETR model incorporates a lightweight FasterNet backbone and a Convolutional and Attention Fusion Module (CAFM).
- A DRB-HSFPN feature pyramid network enhances multi-scale feature integration, and an Inner-EIoU loss function improves accuracy and convergence.
- The model demonstrated a 4.5% increase in mAP@50 and a 2.8% increase in mAP@50:95 on the HAM10000 dataset, with a detection speed of 59.1 FPS.
Impact:
- The proposed SD-DETR model significantly enhances skin disease detection performance.
- It effectively extracts and integrates multi-scale features for improved diagnostic accuracy.
- The model achieves substantial reductions in parameter count (46.0%) and computational load (67.2%), making it highly efficient.
