Semi-automatic volumetric measurement of lung cancer using multi-detector CT effects of nodule characteristics

The aims of this study were to determine interobserver variability in volume measurements of lung tumors (including part-solid and nonsolid types) using novel computer-aided diagnosis (CAD) tools and a machine learning approach and to determine the potential reasons for variability.

In 60 consecutive patients with peripheral lung cancer, the three-dimensional volumes of nodules were measured using the perimeter method by an experienced chest radiologist. In addition, for the same patients, maximal diameters and three-dimensional volumes were measured with and without a novel CAD tool by six observers. The coefficient of variance (CV) as index of interobserver variability was calculated. For the measurement of volume, the results of the perimeter method were compared to those of the CAD method. Furthermore, the CV was calculated for the following subgroups: nodule diameter, internal opacity, margin, spiculation, and adherence to vessels and the chest wall.

There was significant interobserver variability among the six observers for manual, but not CAD, measurements of maximal diameter (P < .001 and P = .207, respectively). Volume measured with the perimeter method by a chest radiologist was well correlated with volume measured with the aid of the CAD system by six radiologists (r = 0.98-0.99). There was no significant difference in the CV for size, internal opacity, spiculation of nodules, or adherence to pulmonary vessels and the chest wall. The CV was significantly higher for obscure marginal nodules than for clear marginal nodules (P < .01).

The novel CAD tool could be used to measure the volume of not only solid but also part-solid and nonsolid lung tumors.