BRAIN CT REGISTRATION USING HYBRID SUPERVISED CONVOLUTIONAL NEURAL NETWORK: FEASIBILITY AND RELIABILITY (ID 1570)
Abstract
Background And Aims
Due to the complex morphology of human brain structures and low soft-tissue contrast of Computed Tomography (CT), brain CT registration is still a great challenge. This study developed a HSCN-Net which may be used for assessment of early ischemic change in Acute Ischemic Stroke (AIS) brain CT.
Methods
HSCN-Net generates virtual deformation fields by a simulator to solve the lack of registration gold standards. The simulator are used to generate multi-scale deformation fields to overcome the large deformation challenge. HSCN-Net adopts a hybrid loss function with deformation field and image similarity to improve the registration accuracy and generalization ability(Fig. 1).
Results
One hundred and one brain CT cases were used for HSCN-Net training and evaluation, and the evaluation results were compared with Demons and VoxelMorph. For visual evaluation(Fig. 2), HSCN-Net was similar to Demons, both are superior to VoxelMorph. Moreover, HSCN-Net was more competent for large and smooth deformation. For quantitative evaluation, the Endpoint error (EPE) mean of HSCN-Net (3.29 mm) was lower than that of Demons (3.47 mm) and Voxelmorph (5.12 mm), the DICE mean was 0.96, which was better than that of Demons (0.94) and Voxelmorph (0.89), and the NMI mean(0.83) was slightly lower than that of Demons(0.84) but higher than that of Voxelmorph (0.81). In addition, the registration time (17.86 s) was lower than that of VoxelMorph (18.53 s) and Demons (147.21 s).
Conclusions
This study developed a HSCN-Net method which may be used as assessment of focal hypoattenuation at brain CT in early acute stroke.
This research is supported by the Shenyang Science and Technology Plan Fund (No. 20-201-4-10)
Trial Registration Number
Not applicable