Results
The examples of 2D vs. 3D lung segmentations for 4 cases are depicted in Figure 2; motion corruption of the axial-plane stack can clearly be appreciated in the coronal plane, with the coronal section of the DSVR shown alongside each case. DSVR was able to produce clear images even from severely motion-corrupted stacks (Figure 2C) and the resultant images can be examined in any plane and at a higher resolution than that of the original input stack (Figure 1, Video 1). The resultant segmentation of fetal lungs demonstrates a smooth outline, with normal cases depicting sufficient detail even to appreciate the lingula (Figure 2A) or a systemic feeding vessel in cases of CLL (Figure 2B). Comparison of the volumes generated from 2D and 3D segmentation is made in Figure 3; results correlated strongly, however a consistently lower measurement was made with 3D segmentation (Bland-Altman: Bias -1.44cm3 [95% confidence limits -2.63 to -0.24]). This was felt to be due to the enhanced ability to exclude structures of the vascular hilum in 3D reconstructions, as well as an inherent reduction in interpolation error.
Having defined the fidelity of DSVR-derived 3D TLV, we proceeded to produce a normal curve based upon 100 healthy cases undergoing fetal MRI as normal controls for research, presented in Figure 4. The corresponding fitted TLV normogram has a high correlation with the previously reported gestational TLV curves produced from conventional 2D segmentation data by Cannie et al11 and Meyers et al12. The generated regression line had a similar R2 (0.72) to those published and was subsequently used to generate the “Expected” value for the depiction of Observed:Expected total lung volume in the abnormal cases.
The manually segmented 2D lung volumes and corresponding DSVR lung volumes are depicted graphically in Figure 5 with reference to the DSVR-derived normal curve. Table 1 displays data calculated from anomaly cases. CPAM lesion volumes were calculated from DSVR reconstructions and are presented as a percentage of the remaining TLV; of note one case had CDH and a BPS of the lung on the same side as the hernia, this case had significantly smaller lesion volume related to lung volume.