Interpretable vertebral fracture quantification via anchor-free landmarks localization

Zakharov, Alexey; Pisov, Maxim; Alim, Bukharaev,; Petraikin, Alexey; Морозов, С. П.; Gombolevskiy, Victor A.; Belyaev, Mikhail

doi:10.1016/j.media.2022.102646

Cited by 15 publications

(10 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…These results are in line with the state-of-the-art, where areas of 90-98 % were reported for the vertebra level classifications (on different datasets). 6,9 To display a single ROC-curve for the multiple models (Fig. 3) trained in a 4-fold cross-validation setting, we concatenated the model's outputs (as if they all came from the same model) on the respective validation set and computed the ROC curve using there combined results.…”

Section: Vertebra Level Evaluationmentioning

confidence: 99%

“…While a direct comparison to the state-of-the-art is not possible on patient level due to different tasks and datasets, sensitivities and specificities around 90 % are not uncommon in the fracture detection literature. [5][6][7][8][9]11 The processing time for a new CT image depended on its field of view and resolution but typically took around 10 seconds, which is reasonable for a clinical workflow and much faster than some previous approaches that require around 60 s. 8,9 The code was not optimized for processing speed and the vertebra localization tool-which used most of the processing time-was available as a CPU-based version only.…”

Section: Patient Level Evaluationmentioning

confidence: 99%

“…In all of these approaches, however, fracture grading was reduced to a binary classification problem: Grade {0} vs {1, 2, 3}, 5, 8 grade {1, 2} vs {2, 3}, 6 or even {0} vs {2, 3} excluding grade 1 vertebrae. 7 Only Zakharov et al 9 directly computed a variant of the Genant grade (differing slightly from the original definition 4 ) from the heights measured between six points along the outline of the vertebra as seen on a 2D reconstruction. Because it is not easy to distinguish fracture grades 0 and 1 and the predictive power (for future fractures) of single mild fractures is unclear, the German DVO-guidelines 10 recommend checking if there is at least one moderate (Genant grade 2) or at least two mild (Genant grade 1) fractures as part of the indication for medicinal therapy.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Towards fracture risk assessment by deep-learning-based classification of prevalent vertebral fractures

Yilmaz¹,

Fricke²,

Laue³

et al. 2023

Medical Imaging 2023: Computer-Aided Diagnosis

View full text Add to dashboard Cite

Osteoporosis is a complex multifactorial skeletal disease and has become a major socioeconomic issue, causing tremendous hospitalization and rehabilitation costs. In addition to age, premature menopause or the use of oral corticosteroids, vertebral fractures are regarded as main risk factors for developing osteoporosis and associated fractures. In this work, we adapt an existing automated pipeline to classify fracture grades of individual vertebrae and the spine as a whole: First, vertebral body centers were identified on CT images by a hierarchical neural network. Next, the fracture grades of individual vertebrae were processed by a multi-head, feed-forward convolutional neural network. The sum of the classified grades was then evaluated according to the German radiology guidelines. A hyperparameter search on validation data showed the most promising results for an output configuration based on three sequentially applied binary classification outputs trained using binary cross-entropy: Grade 0-1 vs 2-3, and 0 vs 1-3, and 0-2 vs 3. In a cross-validation setting on 159 low-dose CT images, our pipeline accurately classified patients to have sum-scores ≥ 2 with sensitivities and specificities of 90 % ± 5.0 % and 87 % ± 2.7 %, respectively. As our method was based on classifying individual vertebrae, we were able to provide both the fracture position and severity to enhance transparency, interpretability and usability.

show abstract

Section: Vertebra Level Evaluationmentioning

confidence: 99%

Section: Patient Level Evaluationmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Towards fracture risk assessment by deep-learning-based classification of prevalent vertebral fractures

Yilmaz¹,

Fricke²,

Laue³

et al. 2023

Medical Imaging 2023: Computer-Aided Diagnosis

View full text Add to dashboard Cite

show abstract

“…So far, most deep learning-based approaches for classification of vertebral body fractures are trained on intensity images in an end-to-end manner. Even after the VerSe dataset/ benchmark for segmentation and localization of vertebrae was introduced in 2019 and 2020 [3][4][5], most introduced methods only rely on the localization of vertebrae instead of leveraging the available segmentation masks [6][7][8]. The amount of available ground truth segmentation masks of vertebrae was further increased by the TotalSegmentator dataset [9,10].…”

Section: Introductionmentioning

confidence: 99%

Shape Matters: Detecting Vertebral Fractures Using Differentiable Point-Based Shape Decoding

Hempe,

Bigalke,

Heinrich

2024

Information

View full text Add to dashboard Cite

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 art through robust automatic segmentation models trained without fracture labels and recent geometric deep learning techniques. Our shape auto-encoders are pre-trained on a large set of vertebrae surface patches. This pre-training step addresses the label scarcity problem faced when learning the shape information of vertebrae for fracture detection from image intensities directly. We further propose a novel shape decoder architecture: the point-based shape decoder. Results: Employing segmentation masks that were generated using the TotalSegmentator, our proposed method achieves an AUC of 0.901 on the VerSe19 testset. This outperforms image-based and surface-based end-to-end trained models. Our results demonstrate that pre-training the models in an unsupervised manner enhances geometric methods like PointNet and DGCNN. Conclusion: Our findings emphasize the advantages of explicitly learning shape features for diagnosing osteoporotic vertebrae fractures. This approach improves the reliability of classification results and reduces the need for annotated labels.

show abstract

“…We used their medical, genetic, and physical characteristics as well as their laboratory test records as our three data sets from different areas for developing a well‐trained and reliable prediction model. When an automated model, based on medical and technological algorithms, is used to diagnose osteoporotic issues then the risk factors 15 and potential problems can be reduced to half and patients having any genetic bone issue can take some precautionary measures before time.…”

Section: Introductionmentioning

confidence: 99%

Machine learning‐based prediction of osteoporosis in postmenopausal women with clinical examined features: A quantitative clinical study

Ullah,

Rehman,

Anwar

et al. 2023

Health Science Reports

View full text Add to dashboard Cite

Osteoporosis is a skeletal disease that is commonly seen in older people but often neglected due to its silent nature. To overcome the issue of osteoporosis in men and women, we proposed an advanced prediction model with the help of machine learning techniques which can help to identify the potential occurrence of this bone disease by its advanced screening tools. To achieve more reliable and accurate results, various machine‐learning techniques were applied to the presented data sets. Moreover, we also compared the performance of our results with other existing algorithms to solely focus on the advanced features of the proposed methodology. The two data sets, the clinical tests of patients in Taiwan and medical reports of postmenopausal women in Korea through Korean Health and Nutrition Examination Surveys (2010–2011) were considered in this study. To predict bone disorders, we utilized the data about females and developed a system using artificial neural networks, support vector machines, and K‐nearest neighbor. To compare the performance of the model Area under the Receiver Operating Characteristic Curve and other evaluation metrics were compared. The achieved results from all the algorithms and compared them with Osteoporosis Self‐Assessment Tool for Asians and the results were noticeably better and more reliable than existing systems due to the involvement of ML. Using machine learning techniques to predict these types of diseases is better because physicians and patients can take early action to prevent the consequences in advance.

show abstract

Interpretable vertebral fracture quantification via anchor-free landmarks localization

Cited by 15 publications

References 25 publications

Towards fracture risk assessment by deep-learning-based classification of prevalent vertebral fractures

Towards fracture risk assessment by deep-learning-based classification of prevalent vertebral fractures

Shape Matters: Detecting Vertebral Fractures Using Differentiable Point-Based Shape Decoding

Machine learning‐based prediction of osteoporosis in postmenopausal women with clinical examined features: A quantitative clinical study

Contact Info

Product

Resources

About