Morphology-Based Individual Vertebrae Classification

Abstract

The human spine is composed, in non-pathological cases, of 24 vertebrae. Most vertebrae are morphologically distinct from the others, such as C1 (Atlas) or C2 (Axis), but some are morphologically closer, such as neighboring thoracic or lumbar vertebrae. In this work, we aim at quantifying to which extent the shape of a single vertebra is discriminating. We use a publicly available MICCAI VerSe 2019 Challenge dataset containing individually segmented vertebrae from CT images. We train several variants of a baseline 3D convolutional neural network (CNN) taking a binary volumetric representation of an isolated vertebra as input and regressing the vertebra class. We start by predicting the probability of the vertebrae to belong to each of the 24 classes. Then we study a second approach based on a two-stage aggregated classification which first identifies the anatomic group (cervical, thoracic or lumbar) then uses a group-specific network for the individual classification. Our results show that: i) the shape of an individual vertebra can be used to faithfully identify its group (cervical, thoracic or lumbar), ii) the shape of the cervical and lumbar seems to have enough information for a reliable individual identification, and iii) the thoracic vertebrae seem to have the highest similarity and are the ones where the network is confused the most. Future work will study if other representations (such as meshes or pointclouds) obtain similar results, i.e. does the representation have an impact in the prediction accuracy?

Vertebrae localization, segmentation and identification using a graph optimization and an anatomic consistency cycle

“…Additionally, it allows to merge duplicate locations with a parameter-free approach. -We extend the work on individual vertebra classification (Mohammed et al, 2020) and show significant improvement on vertebrae classification by including neighboring information. We experimentally quantify that solely using the shape of the vertebrae achieves higher accuracy than using as input the CT image.…”

“…Second, the individual vertebra goes into a per group classifier to predict the identity within the group (N = 7/12/5). As shown in (Mohammed et al, 2020), the two-stage method obtained a higher accuracy than directly classifying into one of the 24 categories.…”

“…This approach reduces the intra-class variance from identifying one out of 24 vertebrae into identifying one out of 7 cervical, 12 thoracic and 5 lumbar vertebrae. Similarly, Mohammed et al (2020) quantify that a twostep approach has a higher identification rate. In our work we leverage this two step approach and extend the work of Mohammed et al (2020) by including neighbouring vertebrae.…”

“…Similarly, Mohammed et al (2020) quantify that a twostep approach has a higher identification rate. In our work we leverage this two step approach and extend the work of Mohammed et al (2020) by including neighbouring vertebrae. Payer et al (2020) use a three-step approach to first localize the spine, then simultaneously locate and identify each vertebra and finally segment each individual vertebra.…”

“…The individual vertebra identification module builds on the work by (Mohammed et al, 2020) where a 3D network is used to classify vertebra into one of their 24 categories. The input of the network is a binary volume of size 128 × 128 × 128 that contains a vertebra segmentation mask.…”