As computer-aided detection continues to make inroads into virtual colonoscopy territory, studies presented Tuesday at the RSNA meeting highlighted its potential for spotting polyps.
As computer-aided detection continues to make inroads into virtual colonoscopy territory, studies presented Tuesday at the RSNA meeting highlighted its potential for spotting polyps.
Using CAD software developed by the National Institutes of Health, researchers at the NIH, the National Naval Medical Center, Walter Reed Army Medical Center, and the Uniformed Services University of Health Sciences examined patient data from 1186 colon cancer screening patients at three medical centers.
The patient population contained two cancers and 180 adenomas 8 mm or larger, according to lead author Dr. Ronald Summers, chief of clinical image processing service and virtual endoscopy and computer-aided diagnosis at the NIH.
The software achieved per polyp sensitivities of 76.4% for polyps 8 mm or larger and 96.2% for polyps 10 mm or larger. Per patient sensitivities were 85.4% and 89.3%, respectively. Summers reported false-positive rates of 6.7 per patient for the smaller polyps and 2.1 for the larger ones.
CAD detected both cancers and exhibited a 0.7 false-positive rate per patient.
"Our CAD systems had high performance comparable to optical colonoscopy for polyps greater than or equal to 1 cm," Summers said.
The NIH CAD program was compared with the Siemens Medical Systems Polyp-Enhanced Viewing system in a study presented by Dr. Joel Fletcher, an assistant professor of radiology at the Mayo Clinic in Rochester, MN.
Fletcher and colleagues compared the two proprietary systems, using a library of 65 clinical CT colonography data sets. All of the data sets had colonoscopy correlation.
The Siemens CAD exhibited a 61% sensitivity for detecting lesions 1 cm or larger compared with a 96% sensitivity for the NIH system. However, the Siemens CAD showed a better specificity, with one false positive per case compared with 4.6 per case for the NIH CAD.
The performances of the two systems were similar to two different radiologists who may be attuned to either a higher specificity or a higher sensitivity. The performance characteristics of a CAD system will usually depend on what a radiologist wants to do - use the system as a first read or check on a primary read, Fletcher said.
The Siemens system got a second workout in a study led by Dr. Anno Graser from the University of Munich, Grosshadern. Graser and colleagues tested it in 105 cases. The CAD system in this study exhibited a 2.2 false-positive rate per data set. It had a 94% sensitivity for medium polyps (7 to 9 mm) and 87% for polyps 10 mm or larger, compared with 94% and 93%, respectively, for expert readers.
Graser noted that there may have been a selection bias in the study, as an attending radiologist selected the cases and may have chosen cases with more readily observable polyps. This could explain the discrepancy between the sensitivity achieved by the CAD system in his study compared with the study presented by Fletcher, he said.
"CAD shows promising sensitivity in the detection of clinically significant polyps but does not outperform expert readers," Graser said.
European Society of Breast Imaging Issues Updated Breast Cancer Screening Recommendations
April 24th 2024One of the recommendations from the European Society of Breast Imaging (EUSOBI) is annual breast MRI exams starting at 25 years of age for women deemed to be at high risk for breast cancer.
New Research Examines Socioeconomic Factors with Mammography No-Shows
April 10th 2024Patients with Medicaid or means-tested insurance were over 27 percent more likely to miss mammography appointments, and only 65 percent of women with three of more adverse social determinants of health had a mammography exam in a two-year period covering 2020 and 2021, according to new research and a report from the CDC.
Mammography Study: AI Improves Breast Cancer Detection and Reduces Reading Time with DBT
April 3rd 2024An emerging artificial intelligence (AI) model demonstrated more than 12 percent higher specificity and reduced image reading time by nearly six seconds in comparison to unassisted radiologist interpretation of digital breast tomosynthesis (DBT) images.