• AI
  • Molecular Imaging
  • CT
  • X-Ray
  • Ultrasound
  • MRI
  • Facility Management
  • Mammography

Diagnosing Pancreatic Lesions on Abdominal CT: Study Says Deep Learning System is Comparable to Radiologist Assessment

Article

Researchers showed the deep learning system had an area under the curve (AUC) ranging from 87 percent to 91 percent in two test sets for diagnosing solid pancreatic lesions of any size and cystic lesions 1 cm or larger on high-contrast computed tomography (CT).

Given the increasing imaging volume in radiology and a recent recognition that a radiologist’s experience level factors into the accuracy of detecting pancreatic lesions on computed tomography (CT), the emergence of a new deep learning system may facilitate improved efficiency and accuracy of assessing these lesions.1,2

In a retrospective study recently published in Radiology, researchers examined the effectiveness of a deep learning system for diagnosing solid and cystic pancreatic lesions on abdominal CT. Trained with CT data from 852 patients, including those who had resection of pancreatic lesions and patients who had no pancreatic lesions, the deep learning system was subsequently evaluated in two test sets of 603 patients (median age of 58) and 589 patients (median age of 63), according to the study.

The study authors found the deep learning system demonstrated a 98 to 100 percent sensitivity rate for detecting solid pancreatic lesions of any size in comparison to 95 to 100 percent sensitivity for radiologist assessment. For cystic lesions 1.0 cm or larger, the deep learning system had a 92 to 93 percent sensitivity rate in comparison to a 93 to 98 percent sensitivity rate for radiologist assessment. The deep learning system had a 91 percent area under the curve (AUC) for the first test set and an 87 percent AUC for the second test set, according to the study authors.

“The (deep learning)-based approach demonstrated overall high diagnostic performance in identifying patients with solid or cystic pancreatic lesions in both test sets, suggesting that (deep learning) has the potential to be used as a supportive tool in diagnosing pancreatic lesions encountered at abdominal CT,” wrote Hyoung Jung Kim, M.D., Ph.D, who is affiliated with the Department of Radiology and the Research Institute of Radiology with the University of Ulsan College of Medicine and Asan Medical Center in Seoul, Korea, and colleagues.

The deep learning system had sensitivity rates that were equivalent to those of radiologists for a variety of pancreatic lesions including intraductal papillary mucinous neoplasms and serous cystic neoplasms. For unspecified benign cystic lesions, the sensitivity rate for the deep learning system was significantly lower in the test sets (54 to 61 percent) than that of the two reviewing radiologists (75 to 93 percent). The deep learning system also had a lower sensitivity rate (16 to 38 percent) for sub-centimeter pancreatic lesions, according to the study authors.

In regard to study limitations, the study authors noted the absence of reference data led to the exclusion of many patients from the initial study sample and this may have caused selection bias with the study. The lack of fully consecutive patient recruitment for the test sets may also have contributed to potential selection bias, according to Kim and colleagues. While the deep learning system provides detection and segmentation of pancreatic lesions, the authors point out the system does not provide lesion classification.

References

1. Rhee H, Park MS. The role of imaging in current treatment strategies for pancreatic adenocarcinoma. Korean J Radiol. 2021;22(1):23-40.

2. Park HJ, Shin K, You MW, et al. Deep learning-based detection of solid and cystic pancreatic neoplasms at contrast-enhanced CT. Radiology. 2022 Aug 23;220171. doi: 10.1148/radiol.220171. Online ahead of print.

Recent Videos
Can Fiber Optic RealShape (FORS) Technology Provide a Viable Alternative to X-Rays for Aortic Procedures?
Does Initial CCTA Provide the Best Assessment of Stable Chest Pain?
Nina Kottler, MD, MS
Practical Insights on CT and MRI Neuroimaging and Reporting for Stroke Patients
Related Content
© 2024 MJH Life Sciences

All rights reserved.