A CAD utilizing 3D massive-training ANNs for detection of flat lesions in CT colonography: preliminary results

Abstract

ABSTRACT Our purpose was to develop a computer-aided diagnostic (CAD) scheme for detection of flat lesions (also known as superficial elevated or depressed lesions) in CT colonography (CTC), which utilized 3D massive-training artificial neural networks (MTANNs) for false-pos itive (FP) reduction. Our CAD scheme c onsisted of colon segmentation, polyp candidate detection, linear discriminant analysis, and MTANNs. To detect flat lesions, we developed a precise shape analysis in the polyp detection step to accommodate the analysis to include a flat shape. With our MTANN CAD scheme, 68% (19/28) of flat lesions, including six lesions “missed” by radiologists in a multicenter clinical trial, were detected correctly, with 10 (249/25) FPs per patient. Keywords: flat neoplasm, superficial elevated lesions, virtual colonoscopy, computer-aided diagnosis, missed polyps, massive training, artificial neural networks 1. INTRODUCTION Colorectal cancer is the second leading cause of cancer deaths in the U.S. CT colonography (CTC) is a technique for detecting colorectal neoplasms by use of a CT scan of the colon. The diagnostic performance of CTC in detecting polyps (i.e., precursors of cancer), however, remains uncertain because of a propensity for perceptual errors. Computer-aided detection (CAD) of polyps has the potential to overcome this difficulty with CTC. A major challenge in CAD development is the detection of flat lesions (also known as flat polyps, superficial elevated or depressed lesions/neoplasms), because existing CAD schemes are designed for detecting the common bulbous polyp shape. Flat lesions in the colon are a major source of false-negative interpretations in CTC [1]; thus, detection of flat lesions by CAD is very important. Because flat lesions are also a source of false negatives in “reference-standard” optical colonoscopy [1], the detection of flat lesions would impact on colorectal cancer mortality. Some flat lesions are known to be histologically aggressive; therefore, the detection of su ch lesions is critical clinically [1,2]. One study showed that flat polyps contributed to 54% of superficial carcinomas [3]. Our purpose was to develop a CAD scheme for detection of flat lesions in CTC, which utilized 3D massive-training artificial neural networ ks (MTANNs) for false-positive (FP) reduction.