Gastric cancer detection using MDCT compared with 2D axial CT: diagnostic accuracy of three different reconstruction techniques.

Abstract

To assess the diagnostic accuracy of different reconstruction techniques using MDCT for gastric cancer detection compared with 2D axial CT.
During 7 months, we performed CT examinations of 104 consecutive patients with gastric cancer and of a control group composed of 35 patients without gastric disease. All gastric cancer was pathologically proven by endoscopy and surgery. Among 104 patients with gastric cancer, 63 patients had early gastric cancer (EGC). Two radiologists retrospectively and independently interpreted the axial CT and three different reconstruction techniques including multiplanar reformation (MPR), transparent imaging (TI), and virtual gastroscopy (VG), using a commercially available, 3D workstation. They graded the presence or absence of gastric cancer in each image sets using a five-point scale and, if present, they assessed its location. Diagnostic accuracy was compared using the area under the receiver operating characteristic curve (Az) for both gastric cancer and only EGC. Sensitivity and specificity were also calculated for each image technique. The k statistics were used to determine inter-observer agreement.
The diagnostic accuracy for overall gastric cancer detection for each of the image sets was as follows: 2D axial CT (Az = 0.858); MPR (Az = 0.879); TI (Az = 0.873); and VG (Az = 0.928). VG had significantly better performance than 2D axial CT (p = 0.016). The sensitivity and specificity were as follows: 76.7% and 82.9% in axial CT; 79.6% and 85.7% in MPR; 91.3% and 80% in TI; and 95.1% and 74.3% in VG. In EGC, the diagnostic performance for its detection was as follows: axial CT (Az = 0.777); MPR (Az = 0.811); TI (Az = 0.825); and VG (Az = 0.896). VG had significantly better performance than both 2D axial CT (P = 0.006) and MRP (P = 0.038). The sensitivity and specificity were as follows: 62.9% and 82.9% in axial CT; 67.7% and 85.7% in MPR; 85.5% and 80% in TI; and 91.9% and 74.3% in VG. The inter-observer agreement showed substantial agreement (κ = 0.67-0.75).
Among the different reconstruction techniques, VG accurately detects gastric cancer and is especially useful for EGC compared with 2D axial CT.