Detecting, Identifying, Grouping and Analysing All Existing Diseases in Large-Scale Chest CT Database with Deep Learning
diseases are there in the chest area and how is the corresponding prevalence. We plan to achieve this goal with minimum manual efforts
and this is supported by deep learning techniques we developed for automatic chest CT analysis. With deep learning models, we plan 1)
apply a disease detection model on CT images to localize suspected regions with diseases, 2) apply a disease identification model onto
the localized regions to assign each region with specific disease type and filter out false positives, 3) apply an image-based clustering
model to separate identified regions into groups, and 4) perform longitudinal analysis and extract some information to provide potential
evidence for clinical evaluation. Thereafter, if needed, we will recruit doctors to verify our generated disease groups. Lastly, we
summarize the disease types and prevalence according to the above findings.
1. Develop a disease detection model to detect disease regions from chest CT images of the NLST database. We aim to find a variety of
lesions both inside and outside the lungs, not limited to lung nodules.
2. Track the lesions’ changes in multiple CT scans of the same patient for longitudinal analysis.
3. Develop a disease identification model to automatically assign each detected disease region with a disease type.
4. Develop a disease clustering model to separate the identified diseases into groups.
5. Find some potential evidence for clinical evaluation based on longitudinal analysis.
6. To summarize types and prevalence of diseases in the NLST database.
Jinzheng Cai, PAII Inc.
Ke Yan, PAII Inc.
Youbao Tang, PAII Inc.