Automated scoring of aortic calcification in vertebral fracture assessment images

Abstract

The severity of abdominal aortic calcification (AAC) is a strong, independent predictor of cardiovascular disease (CVD). Vertebral fracture assessment (VFA) is a low radiation screening tool which can be used to incidentally measure AAC. This work compares the performance of Haar feature random forest classification with a Unet based convolutional neural network (CNN) segmentation, to automatically quantify AAC. Clinical semiquantitative scores were also generated using U-net. Scores were calculated using the relative length of labelled calcification and compared to manual scoring. The U-net outperformed the random forest, showed sensible segmentations and AAC scores, though it could not match human annotation accuracy.

Automatic Segmentation and Quantification of Abdominal Aortic Calcification in Lateral Lumbar Radiographs Based on Deep-Learning-Based Algorithms

“…Our study has improved the U-Net model accuracy by using a larger training dataset and producing continuous AAC score measures. This technique yielded an R 2 coefficient of 0.82, surpassing the 0.49 obtained by Petersen et al [15] and the 0.59 obtained by Luke et al [28] for lateral radiographs. Furthermore, another study by Fusaro et al [32] proposed a semiautomatic tool for quantifying AAC.…”

“…The model was trained on 20 images and tested on another 53 images. Luke et al [28] used 1600 dual-energy X-ray absorptiometry images to automatically calculate the AAC score, of which only 195 images containing evidence of AAC were included for the model development. They compared the performances of random forest classification and a U-Net model with that of human annotation.…”

“…The U-shaped fully convolutional network (U-Net) architecture [23] and its variants have excelled in segmentation tasks [24][25][26][27]. Several studies have demonstrated that deep learning techniques outperform machine learning techniques in automatically scoring aortic calcification [12,28].…”

Show, Attend and Detect: Towards Fine-Grained Assessment of Abdominal Aortic Calcification on Vertebral Fracture Assessment Scans

“…They used non-standard class boundaries to classify the test images with an average accuracy of 92.9%. Chaplin et al [3] followed a similar process as [4] on 195 DXA VFA scans to extract the Region of Interest (ROI), except that, they used a statistical shape model. The ROI in each scan was then warped to straighten the spine which generally leads to loss of information in a calcified aorta.…”

“…The published methods [4,3,15] predict an overall AAC-24 score for each scan. A brief summary of these methods is give in Section 2.2.…”