DeepLNAnno: a Web-Based Lung Nodules Annotating System for CT Images.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Lung cancer diagnosis relies heavily on accurate pulmonary nodule detection from CT scans.
• Deep neural networks show promise for nodule detection but require extensive, accurately labeled data.
• Existing annotation systems are suboptimal for pulmonary nodule labeling in CT imaging.

Purpose of the Study:

• To develop and evaluate DeepLNAnno, a web-based system for efficient and accurate annotation of pulmonary nodules in CT images.
• To address the limitations of current annotation tools for deep learning-based lung cancer detection.

Main Methods:

• Development of DeepLNAnno, a three-tier, web-based annotation system with semi-automatic features.
• Collaboration with a medical group from West China Hospital for data annotation.
• Training and evaluation of a deep neural network model using the annotated data.

Main Results:

• DeepLNAnno demonstrated ease of use and improved label accuracy compared to existing systems.
• A substantial dataset of annotated lung CT images was successfully collected.
• The developed nodule-detection system achieved good benchmark scores on evaluation data.

Conclusions:

• DeepLNAnno is an effective tool for pulmonary nodule annotation, facilitating the creation of high-quality datasets.
• The system supports the development of robust deep learning models for lung cancer screening and diagnosis.