HomeAnalysis & opinionComputer Aided Diagnosis in CT Colonography

Computer Aided Diagnosis in CT Colonography

    Analysis by AJ. Aschoff, MS. Juchems


    Colorectal cancer is a major cause of cancer-related deaths.  Various screening tests are available for colorectal cancer, including the faecal occult blood test, sigmoidoscopy and optical colonoscopy, each with its own advantages and limitations.  Colonoscopy is considered the gold standard for screening for colorectal polyps (1-3), and offers high sensitivity and specificity.  Since David Vining first described CT colonography (CTC, also known as “virtual colonoscopy”) in 1990, its potential was quickly evident and it matured to a serious screening alternative for colorectal cancer.
    CT colonography has benefited greatly from developments in CT scanner technology and new methods of image processing.  In an attempt to further improve the performance of CT colonography, a research emphasis has been put on automated polyp identification using computer-aided detection (CAD) methods.  CAD research started in the 1980s even before CT colonography was developed.  It made its way into clinical routine, for example, in mammography, and further CAD development and research are going on for the detection of lung nodules.
    Three different reading paradigms can be applied to CAD in CTC: CAD as “first reader”, as “ concurrent reader,” and as “second reader”.  As a first reader, CAD separates out negative cases before physicians read the cases.  This paradigm bears the risk of missing colorectal lesions.  CAD as concurrent reader presents suspect lesions to the physician, who can then further evaluate the presented lesion and decide whether it is genuine or not.  It is still unclear whether this type of evaluation biases the reader by drawing attention to CAD findings. CAD used as second reader re-reviews CTC cases.  Thus, CAD findings can be used to alter a report if lesions have been missed in the first place.  This approach is considered, at present, to be the most promising with regard to detection rates.