Regarding spine imaging, landmarks localization has indeed a high relevance in the clinical perspective because it allows characterizing spine alignment in term of angles, distances and shape that could be useful e.g. Among this abundance of data, we have large sets of medical images (CT, MRI or radiographs) that can be used for various tasks such as the detection or the classification of pathologies or the localization of relevant anatomical structures and points of interest. Nowadays we are facing an exponential increase in the availability of biomedical data that open different challenges and opportunities in healthcare research 1.

Our model is able to calculate with good accuracy the coordinates of the vertebral corners and has a large potential for improving the reliability and repeatability of measurements in clinical tasks. All the predicted angles were highly correlated with the ground truth, despite non-negligible absolute median errors of 1.84°, 2.43° and 1.98° for L1–L5, L1–S1 and SS respectively. The median localization errors relative to the vertebrae dimensions were 1.98% and 1.68% for x and y coordinates respectively. We evaluated the model both qualitatively and quantitatively on a set of 195 test images. For the localization tasks, we used a differentiable spatial to numerical transform (DSNT) as the top layer. In step 2, we refined the localization using cropped images of a single vertebra as input to another convolutional neural network and we used geometrical transformations to map the corners to the original image. In step 1, we trained 2 Convolutional Neural Networks to identify each vertebra in the image and calculate the landmarks coordinates respectively. We realized a model that consists of 2 steps. For this purpose, we used 10,193 images annotated with the landmarks coordinates as the ground truth. In this work we propose to use Deep Learning to automatically calculate the coordinates of the vertebral corners in sagittal x-rays images of the thoracolumbar spine and, from those landmarks, to calculate relevant radiological parameters such as L1–L5 and L1–S1 lordosis and sacral slope.

X ray images cervical spine