Project description/goals

Lung cancer is one of the most common malignant tumors. Currently, the use of CT scans for lung cancer screening has become a global consensus. However, a large number of pulmonary nodules with uncertain attribute always been detected by the CT screening. This project aims to carry out intelligent analysis of pulmonary nodules based on lung CT images, focusing on fine-grained analysis of solid components of pulmonary nodules, and constructing an automatic system for pulmonary nodules analysis.

Importance/impact, challenges/pain points

Due to the excessive detection of lung nodules in CT screening and ambiguous difference among different types of nodule solid components, traditional screening method costs radiologists lots of time and usually leads to missing detections and false positives. Automatic intelligent system for lung nodules analysis can greatly reduce the diagnosis time of radiologists and improve the accuracy of screening.

Solution description

We collected a large dataset of CT images with pulmonary nodules, cleaned and screened the images and annotations in cooperation with radiologists to improved the annotation accuracy of existing data.

To solve the problem of ambiguous annotations of solid components, we proposed a new classification framework to make more effective use of annotation information and reduce the negative impact of the mislabeling on the model.

Key contribution/commercial implication

CT screening of lung cancer has a widespread application in clinical practice. Automatic intelligent system for lung nodules analysis can effectively reduce the diagnosis time of radiologists and improve the screening accuracy of lung cancer, which has a wide application prospect.

Next steps

Lung adenocarcinoma is one of the most common types of lung cancer. Pathologically, the development of early lung adenocarcinoma is a continuous and dynamic evolution process, and there exists significant differences in follow-up strategies and treatment methods for different stages of lung adenocarcinoma. Therefore, it is of great significance to distinguish the evolution stages of lung adenocarcinoma for the diagnosis and treatment of lung cancer.  Based on the existing research work on intelligent analysis of solid components of pulmonary nodules, our team is cooperating with Cancer Hospital Chinese Academy of Medical Sciences, Shenzhen Center and working on a research project about quantitative visualization research of molecular pathological information and dynamic evolution of early lung adenocarcinoma.  

Collaborators/partners

Cancer Hospital Chinese Academy of Medical Sciences, Shenzhen Center.

Team/contributors

Xiaoguang Han, Changmiao Wang, Luyue Shi, etc.