Is Visual Registration Equivalent to Semiautomated Registration in Prostate Biopsy?

Abstract

In magnetic resonance iimaging- (MRI-) ultrasound (US) guided biopsy, suspicious lesions are identified on MRI, registered on US, and targeted during biopsy. The registration can be performed either by a human operator (visual registration) or by fusion software. Previous studies showed that software registration is fairly accurate in locating suspicious lesions and helps to improve the cancer detection rate. Here, the performance of visual registration was examined for ability to locate suspicious lesions defined on MRI. This study consists of 45 patients. Two operators with differing levels of experience (<1 and 18 years) performed visual registration. The overall spatial difference by the two operators in 72 measurements was 10.6 ± 6.0 mm. Each operator showed a spatial difference of 9.4 ± 5.1 mm (experienced; 39 lesions) and 12.1 ± 6.6 mm (inexperienced; 33 lesions), respectively. In a head-to-head comparison of the same 16 lesions from 12 patients, the spatial differences were 9.7 mm ± 4.9 mm (experienced) and 13.4 mm ± 7.4 mm (inexperienced). There were significant differences between the two operators (unpaired,Pvalue = 0.042; paired,Pvalue = 0.044). The substantial differences by the two operators suggest that visual registration could improperly and inaccurately target many tumors, thereby potentially leading to missed diagnosis or false characterization on pathology.