Figure - available from: Information
This content is subject to copyright.
To determine the most suitable architecture for our task, we employ combinations of several encoder–decoder architectures including traditional convolutional methods and geometric methods. The AEs are trained to reconstruct either a point cloud representation or a volumetric surface representation of vertebrae, which are derived from the previously computed segmentation mask. As Shape Encoder (A), we employ a convolutional method, as well as a point-based and a graph-based method to predict the embedding z. As Shape Decoder (B), we employ a convolutional method as well as a point-based method and propose a novel point-based shape decoder. The Shape Classifier (C) is then trained separately on the embedding z for each encoder–decoder combination using the same multilayer perceptron (MLP) model. Note that only the weights of the MLP are trained in a supervised manner, whereas the weights of the encoder are fixed.
Source publication
Background: Degenerative spinal pathologies are highly prevalent among the elderly population. Timely diagnosis of osteoporotic fractures and other degenerative deformities enables proactive measures to mitigate the risk of severe back pain and disability. Methods: We explore the use of shape auto-encoders for vertebrae, advancing the state of the...
Similar publications
Osteoporotic vertebral compression fractures, often resulting from low-energy trauma, markedly impair the quality of life of elderly individuals. The present retrospective study focused on the clinical efficacy of unilateral percutaneous vertebroplasty (PVP) in the treatment of osteoporotic compression fractures. A total of 68 patients, representin...
Citations
... Radi-ologists use the Genant scale (Genant et al., 1993) to measure fracture severity from CT images. Deep Learning can automate VCF detection (Valentinitsch et al., 2019;Tomita et al., 2018;Chettrit et al., 2020;Husseini et al., 2020b;Yilmaz et al., 2021;Engstler et al., 2022;Windsor et al., 2022;Iyer et al., 2023;Hempe et al., 2024), however, only a few works have considered VCF grading, all fully-supervised (Pisov et al., 2020;Zakharov et al., 2023;Wei et al., 2022;Yilmaz et al., 2023). Compared to fracture detection, grading is an even more imbalanced task since medium to severely fractured vertebrae account for only a small portion of overall data. ...
... Compared to fracture detection, grading is an even more imbalanced task since medium to severely fractured vertebrae account for only a small portion of overall data. Closest to our approach, Husseini et al. (2020a) and Hempe et al. (2024) train auto-encoders for vertebra shape reconstruction and then use the learned latent codes for downstream fracture detection. ...
Counterfactual explanations (CEs) aim to enhance the interpretability of machine learning models by illustrating how alterations in input features would affect the resulting predictions. Common CE approaches require an additional model and are typically constrained to binary counterfactuals. In contrast, we propose a novel method that operates directly on the latent space of a generative model, specifically a Diffusion Autoencoder (DAE). This approach offers inherent interpretability by enabling the generation of CEs and the continuous visualization of the model’s internal representation across decision boundaries. Our method leverages the DAE’s ability to encode images into a semantically rich latent space in an unsupervised manner, eliminating the need for labeled data or separate feature extraction models. We show that these latent representations are helpful for medical condition classification and the ordinal regression of severity pathologies, such as vertebral compression fractures (VCF) and diabetic retinopathy (DR). Beyond binary CEs, our method supports the visualization of ordinal CEs using a linear model, providing deeper insights into the model’s decision-making process and enhancing interpretability. Experiments across various medical imaging datasets demonstrate the method’s advantages in interpretability and versatility. The linear manifold of the DAE’s latent space allows for meaningful interpolation and manipulation, making it a powerful tool for exploring medical image properties. Our code is available at https://github.com/matanat/dae_counterfactual</a
... Radi-ologists use the Genant scale (Genant et al., 1993) to measure fracture severity from CT images. Deep Learning can automate VCF detection (Valentinitsch et al., 2019;Tomita et al., 2018;Chettrit et al., 2020;Husseini et al., 2020b;Yilmaz et al., 2021;Engstler et al., 2022;Windsor et al., 2022;Iyer et al., 2023;Hempe et al., 2024), however, only a few works have considered VCF grading, all fully-supervised (Pisov et al., 2020;Zakharov et al., 2023;Wei et al., 2022;Yilmaz et al., 2023). Compared to fracture detection, grading is an even more imbalanced task since medium to severely fractured vertebrae account for only a small portion of overall data. ...
... Compared to fracture detection, grading is an even more imbalanced task since medium to severely fractured vertebrae account for only a small portion of overall data. Closest to our approach, Husseini et al. (2020a) and Hempe et al. (2024) train auto-encoders for vertebra shape reconstruction and then use the learned latent codes for downstream fracture detection. ...
Counterfactual explanations (CEs) aim to enhance the interpretability of machine learning models by illustrating how alterations in input features would affect the resulting predictions. Common CE approaches require an additional model and are typically constrained to binary counterfactuals. In contrast, we propose a novel method that operates directly on the latent space of a generative model, specifically a Diffusion Autoencoder (DAE). This approach offers inherent interpretability by enabling the generation of CEs and the continuous visualization of the model's internal representation across decision boundaries. Our method leverages the DAE's ability to encode images into a semantically rich latent space in an unsupervised manner, eliminating the need for labeled data or separate feature extraction models. We show that these latent representations are helpful for medical condition classification and the ordinal regression of severity pathologies, such as vertebral compression fractures (VCF) and diabetic retinopathy (DR). Beyond binary CEs, our method supports the visualization of ordinal CEs using a linear model, providing deeper insights into the model's decision-making process and enhancing interpretability. Experiments across various medical imaging datasets demonstrate the method's advantages in interpretability and versatility. The linear manifold of the DAE's latent space allows for meaningful interpolation and manipulation, making it a powerful tool for exploring medical image properties. Our code is available at https://github.com/matanat/dae_counterfactual.
