Meeting Abstracts
» Identification of Pulmonary Nodules on Chest CT Scans by a Computer-aided Detection (CAD) Algorithm

PURPOSE: To assess the sensitivity and false positive rate of CAD for detection of ≥3mm pulmonary nodules in MDCT scans.

METHODS AND MATERIALS: A two-stage CAD algorithm was developed that differentiates pulmonary nodules from other lung structures based upon their globular geometry and then reduces the number of false positive (FP) identifications by analyzing the shape of the nodule candidates. We applied our CAD algorithm to thoracic MDCT scans (4 x 1.25 mm, 6.0 pitch) from 21 patients with suspicion of pulmonary nodules on CXR. Seven of the 21 patients had no nodules, and the remaining 14 patients had a total of 100 pulmonary nodules ≥3mm (8 >10 mm; 42 6-10 mm; 48 3-6 mm). Two thoracic radiologists, who were blinded to CAD results, determined truth by consensus. All nodule candidates detected by CAD were assigned as being true or false positive results, and false negative results were tallied. By varying CAD performance thresholds, the relationship of CAD sensitivity relative to the number of FP nodules per CT scan was determined and stratified by nodule diameter.

RESULTS: Overall CAD detected 96% of all ≥3 mm nodules. The sensitivity of CAD was 100% for ≥ 10 mm nodules with 7.3 FP/scan, 98% for 6-10 mm nodules with 30 FP/scan, and 92% of 3-6mm nodules with 61 FP/scan. At a target of 20 FP/scan sensitivity was 98, 95, and 83% for ≥ 10 mm, 6-10 mm, and 3-5 mm nodules, respectively. All 4 ≥3 mm nodules not detected by CAD were contiguous with the diaphragm. Analysis of 100 randomly selected FPs showed that 14 were actually 3-4mm nodules that were not included in the gold standard.

CONCLUSION: CAD may detect pulmonary nodules from MDCT images with substantially greater sensitivity than previously published lung CAD results, achieving sensitivity and false positive levels similar to those of radiologists.

This page last modified on 3 February 2004 by Society Webmaster.