Electronic subtraction fools CAD, leads to false positives

Electronically cleansing CT colonography imaging data can digitally remove extraneous fecal matter, making it easier for human observers to detect polyps. It also introduces subtraction artifacts, however, leading to a high number of false-positive readings, according to a study presented at the RSNA meeting.

Computer-aided detection has demonstrated high sensitivity for polyp detection with standard preparation, according to study presenter Dr. Michael Zalis, an assistant radiologist in abdominal imaging and intervention at Massachusetts General Hospital.

Zalis and colleagues wanted to examine the effect minimal preparation along with electronic cleansing would have on CAD accuracy in detecting colorectal polyps. They used 121 electronically cleansed CT colonography cases to train a CAD system. Two readers then read electronically cleansed cases from 150 subjects at moderate risk for colorectal carcinoma and determined that 26 of them had optimal bowel preparations.

These 26 cases were then reprocessed, either undergoing CAD by itself or electronic cleansing followed by CAD.

Sensitivity for the CAD system was 100%. Looking at the CAD results regarding false positives indicated that without digital subtraction, CAD produced 3.8 false positives per dataset. Once electronic cleansing was introduced, however, the number of false positives seen on CAD jumped to an average of 7.2 per data set, a statistically significant difference.

The main cause of false positives on CAD without the electronic cleansing was uneven stool tagging, according to Zalis. For CAD with electronic cleansing, subtraction artifacts, including artifacts at air-fluid interfaces, and tagged fluid and artifacts with morphology of polyps were major causes of false positives.

While these results indicate the need to modify CAD algorithms to be able to handle digital subtraction techniques, it is important not to overinterpret the results, especially regarding the 100% sensitivity, due to very small cohort sample. A dedicated analysis assessing the effect of poor preparation on CAD data is forthcoming, Zalis said.

For more information from the online Diagnostic Imaging archives:

Careful CT colonography technique avoids pitfalls

Virtual colonoscopy brings opportunity and obligation

Virtual colonoscopy technology makes inroads into imaging practice