Computed tomographic colonography (CTC) is a rapidly developing screening method that could significantly improve colorectal cancer prevention. Clinically effective polyp detection using CTC will likely depend on the development of computer-aided polyp detection (CARD). Our long term goal is to develop robust CARD methods by focusing on computer polyp detection (CPD) and high-performance pattern analysis (HPPA). We claim that clinically effective CAPD with sufficient sensitivity and specificity requires CPD methods that utilize HPPA during development, training, and application. We base this claim oh the following observations: 1) uneven results in CTC using human observers alone indicate that the screening task depends greatly on the protocol and on radiologist experience; 2) the routine acquisition of prone and supine scans is currently underused in CPD due to the correspondence problem; 3) current CPD systems rely on a limited set of features, classifier types, and learning methods, and have generally poor specificity. The complexity of these issues introduces computational demands requiring HPPA.
The specific aims are to: 1. Design and implement a high-performance pattern analysis system for CPD. Computer identification of polyps requires solutions to varied image analysis and optimization problems. Considering the complexity of the colon anatomy, a simple pattern recognition method is unlikely to meet acceptable performance criteria. We will extend our current CPD system by applying more sophisticated pattern analysis algorithms that incorporate classifier ensembles and feature discovery. We will use the classifier ensembles to validate automated feature identification and reduction, develop a confidence metric for identified polyps, and map the CPD system to a computer cluster to meet the processing demands. 2. Optimally combine prone and supine CTC scans. Current CPD/CAPD methods that acquire prone and supine scans do not fully utilize the joint information. Colon segment registration, within a patient study for polyp detection and across patient studies for polyp surveillance, is a challenging problem as the colon undergoes deformation during repositioning. We will develop a non-rigid, hierarchical-based method to register the prone and supine colon lumen positions and jointly incorporate them into the CPD system. Colorectal carcinoma is the second leading most prevelant cause of cancer deaths in the US and first among nonsmokers, yet effective screening could reduce mortality by 50%. However, compliance for recommended screening is relatively low. By being more patient friendly, CTC has the potential to increase compliance and thus have a dramatic effect on early detection and treatment. ? ? ?
| Suh, Jung W; Wyatt, Christopher L (2011) Registration under topological change for CT colonography. IEEE Trans Biomed Eng 58:1403-11 |
| Suh, Jung W; Wyatt, Christopher L (2009) Deformable registration of supine and prone colons for computed tomographic colonography. J Comput Assist Tomogr 33:902-11 |
