INTER-VERTEBRAL DISK MODELLING FROM PAIRS OF SEGMENTED VERTEBRAL MODELS USING TRAINABLE PRE-PROCESSING NETWORKS

We introduce a combined approach for modelling intervertebral disks based on pairs of neighbouring vertebral models segmented from T2-weighted MRI. Our approach incorporates a trainable pre-processing pipeline using FC-ResNets which demonstrates that a low-capacity fully convolutional network (FCN) can be used as a pre-processor to normalize the input MRI data. In the proposed segmentation pipeline, we use FCNs to obtain normalized images, which are then iteratively refined by means of a FC-ResNet to generate a segmentation prediction of the vertebral bodies and pedicles. Based on neighbouring endplates, intervertebral disks are modelled for disk replacement procedures. Clinical experiments on a dataset totalling 43 patients demonstrates promising results in terms of vertebra segmentation and disk extraction. Quantitative comparison to expert MR segmentation yields a surface accuracy of 1.1 ± 0.8mm, while a comparison to CT annotations yielded an accuracy of 2.2 ± 1.4mm. The results illustrate the strong potential and versatility of the pipeline by achieving accurate segmentations which can be used for surgical procedures.