Colon cancer is the second leading cause of cancer deaths for men and women in the United States, even though it could be prevented by early detection and removal of its precursor lesions. Computed tomographic colonography (CTC) could substantially increase the capacity, safety, and patient compliance of colorectal examinations. However, the current standard of cathartic bowel preparation for CTC and optical colonoscopy (OC) is poorly tolerated by patients and has been recognized as a major barrier to colorectal examinations. Our advanced non-cathartic multi-center computer-assisted CTC trial showed that non-cathartic CTC is easily tolerated by patients and that radiologists who use computer-aided detection (CADe) can detect large polyps in size in non-cathartic CTC with high sensitivity, comparable to that of OC. However, SF6-lesions (serrated lesions, flat lesions <3 mm in height, and polyps 6 ? 9 mm in size) were a significant source of false negatives in the trial. The challenges of detection and visualization of these SF6-lesions in non-cathartic CTC are caused largely by the inability of the current single-energy CTC technique to differentiate between soft tissues, fecal tagging, and their partial volumes with lumen air. We propose to employ multi-spectral CTC precision imaging and artificial intelligence (AI) to overcome these inherent limitations of non-cathartic CTC. Our goal in this project is to develop a novel deep-learning AI (DEEP-AI) scheme for multi-spectral multi-material (MUSMA) precision imaging, which will use deep super-learning of high-quality spectral CTC (spCTC) precision images to boost the diagnostic performance of non-cathartic CTC. We hypothesize that (1) high-quality MUSMA precision images can be reconstructed from ultra-low-dose (<1 mSv) spCTC scans, (2) DEEP-AI will yield a detection sensitivity for ?6 mm SF6-lesions comparable to that of OC, and that (3) the use of DEEP-AI as first reader will significantly improve radiologists? detection performance for SF6-lesions and reduce interpretation time compared with unaided reading, and that it will yield a detection accuracy comparable to that of OC.
Our specific aims are (1) to establish a non-cathartic spCTC and MUSMA precision image database, (2) to develop a DEEP-AI Interpretation System for visualization and detection of SF6-lesions, and (3) to evaluate the clinical benefit of the DEEP-AI Interpretation System with non-cathartic spCTC cases. Successful development of the proposed DEEP-AI Interpretation System will substantially improve human readers? performance in the detection of SF6-lesions from non-cathartic CTC examinations that address the problem of patient adherence to colorectal screening guidelines. Such a scheme will make non-cathartic CTC a highly accurate and acceptable screening option for large populations, leading to an increased colorectal screening rate, promoting early diagnosis of colon cancer, and ultimately reducing mortality due to colon cancer.

Public Health Relevance

Although colon cancer is the second leading cause of cancer deaths for men and women in the United States, it could be prevented by early detection and removal of its precursor lesions. Successful development of the proposed deep-learning artificial intelligence scheme will substantially improve human readers? performance in detecting colorectal polyps from non-cathartic CTC examinations that addresses the problem of patient adherence to colorectal screening guidelines. Such a scheme will make non-cathartic CTC a highly accurate and acceptable screening option for large populations, leading to an increased colorectal screening rate, promoting early diagnosis of colon cancer, and ultimately reducing mortality due to colon cancer.

Agency
National Institute of Health (NIH)
Institute
National Cancer Institute (NCI)
Type
Research Project (R01)
Project #
5R01CA212382-04
Application #
10054168
Study Section
Biomedical Computing and Health Informatics Study Section (BCHI)
Program Officer
Redmond, George O
Project Start
2017-12-01
Project End
2022-11-30
Budget Start
2020-12-01
Budget End
2021-11-30
Support Year
4
Fiscal Year
2021
Total Cost
Indirect Cost
Name
Massachusetts General Hospital
Department
Type
DUNS #
073130411
City
Boston
State
MA
Country
United States
Zip Code
02114
Tachibana, Rie; Näppi, Janne J; Ota, Junko et al. (2018) Deep Learning Electronic Cleansing for Single- and Dual-Energy CT Colonography. Radiographics 38:2034-2050