Fanjie Lin

Ph.D.

The First Affiliated Hospital of Guangzhou Medical University

None

NLST (Learn more about this study)

NLST-1167

Jan 2, 2024

Community Early Detection of Chronic Obstructive Pulmonary Diseases using Quantitative LDCT Imaging and Machine Learning

Underdiagnosis of chronic obstructive pulmonary disease (COPD) patients is a worldwide problem. According to the China Pulmonary Health study, the prevalence among individuals aged 40 years or older in the general population is 13.7% and nearly 99·9 million adults aged 20 years or older had spirometry-defined COPD. Furthermore, only 12% of individuals had a previous spirometry-defined COPD diagnosis before the survey, and 60.2% of spirometry-defined COPD patients existed little respiratory symptoms. However, although spirometry is considered a “gold standard” for COPD diagnosis, it is often underutilized in primary care settings. Its common causes include lack of expertise in performing spirometry, high cost and time consumption of spirometry, and low confidence in spirometry interpretation. Questionnaire is an effective modality of early screening for COPD, but its generalization may be limited owing to its highly subjective, and distinctions in race, region, economic level, as well as other factors involved in its development. As is known to all, the early diagnosis of COPD is indispensable for the timely initiation of appropriate lifestyle and therapeutic intervention to reduce the prevalence rate, and there is suggestive evidence that early intensive intervention can efficiently delay the deterioration of disease and improve the patient’s quality of life. Therefore, developing new early screening protocols to provide precise detection and evaluation of COPD for optimal clinical decision-making is urgent.
In recent years, artificial intelligence (AI) has developed vigorously, and gradually affected the development of the medical industry, becoming an vital factor in enhancing the medical level. Due to the advantages of computed tomography (CT), such as high density resolution, non duplication of tissue structure images, and the possibilities of multiple imaging post, the quantitative analysis with CT imaging has shown broad application space and enormous potential in COPD early screening and diagnosis. With the widespread utilize of low-dose CT (LDCT) for lung cancer screening, and the opportunistic CT-COPD screening proposed by 2024 GOLD guidance, early detection for COPD patients using these scans can be implemented in community.
However, it cannot be overlooked that published programs are either predominantly based on datasets from Western smokers or focus on patients with moderate or severe COPD in China, who are significantly heterogeneous compared to undiagnosed patients in our communities. Thus, whether those approaches could be applied to distinct datasets extracted from real-world scenarios, especially among the Chinese population, remains largely unknown.
Our team, centered on the community population in Guangdong, Guangzhou, China, has verified the benefits of weakly labeled LDCT images for COPD detection using multimodal data of chest LDCT quantitative characteristics, questionnaires, and CT reports, and successfully constructed a robust machine learning model. We hope that this tool can be applied on a wider scale and specifically propose to use COPD and control LDCT images from NLST project for external validation.