nach oben

La radiologia medica

Erschienen in:

24.02.2025 | Original article

Shapley-based saliency maps improve interpretability of vertebral compression fractures classification: multicenter study

verfasst von: Liang Xia, Jun Zhang, Zhipeng Liang, Jun Tang, Jianguo Xia, Yongkang Liu

Erschienen in: La radiologia medica | Ausgabe 3/2025

Einloggen, um Zugang zu erhalten

Abstract

Purpose

Evaluate the classification performance and interpretability of the Vision Transformer (ViT) model on acute and chronic vertebral compression fractures using Shapley significance maps.

Materials and Methods

This retrospective study utilized medical imaging data from December 2018 to December 2023 from three hospitals in China. The study included 942 patients, with imaging data comprising X-rays, CTs, and MRIs. Patients were divided into training, validation, and test sets with a ratio of 7:2:1. The ViT model variant, SimpleViT, was fine-tuned on the training dataset. Statistical analyses were performed using the PixelMedAI platform, focusing on metrics such as ROC curves, sensitivity, specificity, and AUC values, with statistical significance assessed using the DeLong test.

Results

A total of 942 patients (mean age 69.17 ± 10.61 years) were included, with 1076 vertebral fractures analyzed (705 acute, 371 chronic). In the test set, the ViT model demonstrated superior performance over the ResNet18 model, with an accuracy of 0.880 and an AUC of 0.901 compared to 0.843 and 0.833, respectively. The use of ViT Shapley saliency maps significantly enhanced diagnostic sensitivity and specificity, reaching 0.883 (95% CI: 0.800, 0.963) and 0.950 (95% CI: 0.891, 1.00), respectively.

Conclusion

In vertebral compression fractures classification, Vision Transformer outperformed Convolutional Neural Network, providing more effective Shapley-based saliency maps that were favored by radiologists over GradCAM.

Nur mit Berechtigung zugänglich

Soffer S, Ben-Cohen A, Shimon O et al (2019) Convolutional neural networks for radiologic images: a radiologist’s guide. Radiology 290(3):590–606. https://doi.org/10.1148/radiol.2018180547CrossRefPubMed

Azulay A, Weiss Y (2019) Why do deep convolutional networks generalize so poorly to small image transformations? J Mach Learn Res 20(184):1–25

Liang Y, Li S, Yan C et al (2021) Explaining the black-box model: a survey of local interpretation methods for deep neural networks. Neurocomputing 419:168–182. https://doi.org/10.1016/j.neucom.2020.08.011CrossRef

Dosovitskiy A, Beyer L, Kolesnikov A et al. (2020) An image is worth 16 x 16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929, https://doi.org/10.48550/arXiv.2010.11929.

Beyer L, Zhai X, Kolesnikov A (2022) Better plain vit baselines for imagenet-1k. arXiv preprint arXiv:2205.01580, https://doi.org/10.48550/arXiv.2205.01580.

Cubuk E D, Zoph B, Shlens J et al. (2019) Randaugment: practical data augmentation with no separate search. arXiv preprint arXiv:1909.13719, 2(4): 7. https://doi.org/10.48550/arXiv.1909.13719.

Zhang H, Cisse M, Dauphin Y N et al. (2017) mixup: beyond empirical risk minimization. arXiv preprint arXiv:1710.09412, https://doi.org/10.48550/arXiv.1710.09412.

Xu F, Xiong Y, Ye G et al (2023) Deep learning-based artificial intelligence model for classification of vertebral compression fractures: a multicenter diagnostic study. Front Endocrinol 14:1025749. https://doi.org/10.3389/fendo.2023.1025749CrossRef

Zuo XH, Zhu XP, Bao HG et al (2018) Network meta-analysis of percutaneous vertebroplasty, percutaneous kyphoplasty, nerve block, and conservative treatment for nonsurgery options of acute/subacute and chronic osteoporotic vertebral compression fractures (OVCFs) in short-term and long-term effects. Medicine 97(29):52CrossRef

10.

Hoyt D, Urits I, Orhurhu V et al (2020) Current concepts in the management of vertebral compression fractures. Curr Pain Head Rep 24(52):1–10

11.

Yang H, Yan S, Li J et al (2022) Prediction of acute versus chronic osteoporotic vertebral fracture using radiomics-clinical model on CT. Eur J Radiol 149:110197. https://doi.org/10.1016/j.ejrad.2022.110197CrossRefPubMed

12.

Zhang J, Liu F, Xu J et al (2023) Automated detection and classification of acute vertebral body fractures using a convolutional neural network on computed tomography. Front Endocrinol 14:1132725. https://doi.org/10.3389/fendo.2023.1132725CrossRef

13.

Dong Q, Luo G, Lane NE et al (2022) Deep learning classification of spinal osteoporotic compression fractures on radiographs using an adaptation of the genant semiquantitative criteria. ACAD Radiol 29(12):1819–1832. https://doi.org/10.1016/j.acra.2022.02.020CrossRefPubMedPubMedCentral

14.

Dong Q, Luo G, Lane NE et al (2023) Generalizability of deep learning classification of spinal osteoporotic compression fractures on radiographs using an adaptation of the modified-2 algorithm-based qualitative criteria. ACAD Radiol. 30(12):2973–2987CrossRefPubMedPubMedCentral

15.

Arun N, Gaw N, Singh P et al (2021) Assessing the trustworthiness of saliency maps for localizing abnormalities in medical imaging. Radiol Artif Intell 3(6):e200267. https://doi.org/10.1148/ryai.2021200267CrossRefPubMedPubMedCentral

16.

Eitel F, Soehler E, Bellmann-Strobl J et al (2019) Uncovering convolutional neural network decisions for diagnosing multiple sclerosis on conventional MRI using layer-wise relevance propagation. NeuroImage Clini 24:102003. https://doi.org/10.1016/j.nicl.2019.102003CrossRef

17.

Jin W, Fatehi M, Abhishek K et al (2020) Artificial intelligence in glioma imaging: challenges and advances. J Neural Eng 17(2):021002. https://doi.org/10.1088/1741-2552/ab8131CrossRefPubMed

18.

Zhang J, Petitjean C, Ainouz S (2022) Segmentation-based vs. regression-based biomarker estimation: a case study of fetus head circumference assessment from ultrasound images. J Imag 8(2):23. https://doi.org/10.3390/jimaging8020023CrossRef

19.

Kazawa N (2012) T2WI MRI and MRI-MDCT correlations of the osteoporotic vertebral compressive fractures. Eur J Radiol 81(7):1630–1636. https://doi.org/10.1016/j.ejrad.2011.04.052CrossRefPubMed

20.

Sun Q, Ge Z (2021) A survey on deep learning for data-driven soft sensors. IEEE Trans Ind Inf 17(9):5853–5866CrossRef

21.

Ullah A, Elahi H, Sun Z et al (2022) Comparative analysis of AlexNet, ResNet18 and SqueezeNet with diverse modification and arduous implementation. Arab J Sci Eng 47(2):2397–2417. https://doi.org/10.1007/s13369-021-06182-6CrossRef

22.

Covert I, Kim C, Lee SI (2022) Learning to estimate shapley values with vision transformers. arXiv preprint arXiv:2206.05282, https://doi.org/10.48550/arXiv.2206.05282.

23.

Moujahid H, Cherradi B, Al-Sarem M et al (2022) Combining CNN and GradCAM for COVID-19 disease prediction and visual explanation. Intell Autom Soft Comput 32(2):523CrossRef

24.

Khalid S, Rashwan HA, Abdulwahab S et al (2024) FGR-Net: Interpretable fundus image gradeability classification based on deep reconstruction learning. Expert Syst Appl 238:121644. https://doi.org/10.1016/j.eswa.2023.121644CrossRef

25.

Wollek A, Graf R, Čečatka S et al (2022) Attention-based saliency maps improve interpretability of pneumothorax classification. Radiol Artifi Intell 5(2):e220187. https://doi.org/10.1148/ryai.220187CrossRef

26.

Chefer H, Gur S, Wolf L (2021) Transformer interpretability beyond attention visualization. In: 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp 782–791.https://doi.org/10.48550/arXiv.2012.09838

27.

Jain S, Salman H, Wong E et al. (2022) Missingness bias in model debugging. arXiv preprint arXiv:2204.08945, https://doi.org/10.48550/arXiv.2204.08945

28.

Zhang Y, Khakzar A, Li Y et al (2021) Fine-grained neural network explanation by identifying input features with predictive information. Adv Neural Inf Process Syst 34:20040–20051

29.

Sun X, Xu W (2014) Fast implementation of DeLong’s algorithm for comparing the areas under correlated receiver operating characteristic curves. IEEE Signal Process Lett 21(11):1389–1393. https://doi.org/10.1109/LSP.2014.2337313CrossRef

30.

Alamri F, Dutta A (2021) Multi-head self-attention via vision transformer for zero-shot learning. arXiv preprint arXiv:2108.00045, https://doi.org/10.48550/arXiv.2108.00045

31.

Alzubaidi L, Zhang J, Humaidi AJ et al (2021) Review of deep learning: concepts, CNN architectures, challenges, applications, future directions. J Big Data 8:1–74. https://doi.org/10.1186/s40537-021-00444-8CrossRef

32.

Li Y, Nie J, Chao X (2020) Do we really need deep CNN for plant diseases identification? Comput Electron Agric 178:105803. https://doi.org/10.1016/j.compag.2020.105803CrossRef

33.

34.

Chen W, Liu X, Li K et al (2022) A deep-learning model for identifying fresh vertebral compression fractures on digital radiography. Eur Radiol 5:1–10. https://doi.org/10.1007/s00330-021-08247-4CrossRef

35.

Zhang J, Xia L, Tang J et al (2023) Constructing a deep learning radiomics model based on X-ray images and clinical data for predicting and distinguishing acute and chronic osteoporotic vertebral fractures: a multicenter study. Acad Radiol. https://doi.org/10.1016/j.acra.2023.10.061CrossRefPubMedPubMedCentral

36.

Di Martino F, Delmastro F (2023) Explainable AI for clinical and remote health applications: a survey on tabular and time series data. Artifi Intell Rev 56(6):5261–5315. https://doi.org/10.1007/s10462-022-10304-3CrossRef

37.

Selvaraju R R, CogswelL M, Das A et al. (2024) Grad-CAM: visual explanations from deep networks via gradient-based localization. In: Proceedings of 2017 IEEE International Conference on Computer Vision. Venice: IEEE, 2017, pp 618–626.https://openaccess.thecvf.com/content_ICCV_2017/papers/Selvaraju_Grad-CAM_Visual_Explanations_ICCV_2017_paper.pdf

38.

Jethani N, Sudarshan M, Covert I C et al. (2021) Fastshap: Real-time shapley value estimation. In: International Conference on Learning Representations. https://openreview.net/pdf?id=Zq2G_VTV53T

39.

Adebayo J, Gilmer J, Muelly M et al (2018) Sanity checks for saliency maps. Adv Neural Inf Process Syst 5:359

Titel: Shapley-based saliency maps improve interpretability of vertebral compression fractures classification: multicenter study
verfasst von: Liang Xia
Jun Zhang
Zhipeng Liang
Jun Tang
Jianguo Xia
Yongkang Liu
Publikationsdatum: 24.02.2025
Verlag: Springer Milan
Erschienen in: La radiologia medica / Ausgabe 3/2025
Print ISSN: 0033-8362
Elektronische ISSN: 1826-6983
DOI: https://doi.org/10.1007/s11547-025-01968-2

Importance of technıcal standardızatıon on ultrasound-based lıver fat quantıfıcatıon

Letter to the Editor

The radiologist as an independent “third party” to the patient and clinicians in the era of generative AI

Editorial

Baseline 18F-FDG PET/CT for predicting pathological response to neoadjuvant chemotherapy and prognosis in locally advanced breast cancer patients: analysis of tumor and lymphoid organs metabolic parameters

Diagnostic Imaging in Oncology

Gut-brain axis and neuroplasticity in health and disease: a systematic review

Neuroradiology

Spatial-demographic analysis model for brain metastases distribution

Open Access
Computer Application

Superior performance in classification of breast cancer molecular subtype and histological factors by radiomics based on ultrafast MRI over standard MRI: evidence from a prospective study

Open Access
Breast Radiology

Neu im Fachgebiet Radiologie

Ringen um den richtigen Umgang mit Zufallsbefunden

24.04.2025
DGP 2025
Kongressbericht

Wenn 2026 in Deutschland das Lungenkrebsscreening mittels Low-Dose-Computertomografie (LDCT) eingeführt wird, wird es auch viele Zufallsbefunde ans Licht bringen. Das birgt Chancen und Risiken.

Bald 5% der Krebserkrankungen durch CT verursacht

23.04.2025
Computertomografie
Nachrichten

Die jährlich rund 93 Millionen CTs in den USA könnten künftig zu über 100.000 zusätzlichen Krebserkrankungen führen, geht aus einer Modellrechnung hervor. Damit würde eine von 20 Krebserkrankungen auf die ionisierende Strahlung bei CT-Untersuchungen zurückgehen.

Röntgen-Thorax oder LDCT fürs Lungenscreening nach HNSCC?

06.04.2025
Kopf-Hals-Tumoren
Nachrichten

Personen, die an einem Plattenepithelkarzinom im Kopf-Hals-Bereich erkrankt sind, haben ein erhöhtes Risiko für Metastasen oder zweite Primärmalignome der Lunge. Eine Studie hat untersucht, wie die radiologische Überwachung aussehen sollte.

Statine: Was der G-BA-Beschluss für Praxen bedeutet

06.04.2025
Praxis und Beruf
Nachrichten

Nach dem G-BA-Beschluss zur erweiterten Verordnungsfähigkeit von Lipidsenkern rechnet die DEGAM mit 200 bis 300 neuen Dauerpatienten pro Praxis. Im Interview erläutert Präsidiumsmitglied Erika Baum, wie Hausärztinnen und Hausärzte am besten vorgehen.

Update Radiologie

Bestellen Sie unseren Fach-Newsletter und bleiben Sie gut informiert.

Newsletter bestellen

Springer Medizin

Shapley-based saliency maps improve interpretability of vertebral compression fractures classification: multicenter study