Related Experiment Video
Updated: Jul 15, 2025
07:53
Author Spotlight: Advancing 3D Modeling for Enhanced Diagnosis and Treatment of Pulmonary Nodules in Early-Stage Lung Cancer
Published on: October 13, 2023
1.5K
Improving diagnosis accuracy with an intelligent image retrieval system for lung pathologies detection: a features
Abdelbaki Souid1, Najah Alsubaie2, Ben Othman Soufiene3
1MACS Research Laboratory RL16ES22, National Engineering School of Gabes, Gabes, Tunisia.
Scientific Reports
|October 3, 2023
Summary
This study introduces a novel deep learning system for retrieving similar lung pathology images. The AI model significantly improves accuracy and efficiency in medical image analysis, aiding radiologists.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Science
Background:
- Accurate detection of lung pathologies is crucial for diagnosis, with medical imaging techniques like X-rays and CT scans being vital.
- Deep learning (DL) offers potential in analyzing medical images for lung pathology detection but faces challenges with data requirements and bias.
- Existing methods for retrieving similar annotated medical images can be inefficient and lack precision.
Purpose of the Study:
- To develop an advanced computer-assisted system for automatic retrieval of annotated medical images with similar content.
- To enhance the efficiency and accuracy of feature extraction for medical image analysis using deep learning.
- To provide a tool that assists radiologists in their workflow by facilitating the retrieval of relevant diagnostic images.
Main Methods:
- A novel deep learning-based feature extractor was developed, fusing YOLOv5 for object detection and EfficientNet for noise reduction.
- The system was designed for automatic retrieval of similar annotated chest radiograph images from a large database.
- Rigorous experimentation was conducted on an extensive and diverse dataset of medical images.
Main Results:
- The proposed system achieved a mean average precision (mAP) of 0.488 at a 0.9 threshold, significantly outperforming YOLOv5+ResNet (0.234) and EfficientDet (0.257).
- A substantial precision improvement of approximately 0.352 was observed, reaching 0.864 across all pathologies compared to baseline models.
- The fusion of YOLOv5 and EfficientNet demonstrated superior performance in feature extraction for medical image retrieval.
Conclusions:
- The developed system offers a significant advancement in the automatic retrieval of analogous annotated medical images.
- The fusion of YOLOv5 and EfficientNet provides a powerful and accurate method for feature extraction in medical imaging.
- This research contributes to improving radiologists' workflow efficiency and diagnostic accuracy in identifying lung pathologies.
