Top

European Journal of Nuclear Medicine and Molecular Imaging

Published in:

Open Access 01-03-2022 | Artificial Intelligence | Original Article

Explainable AI to improve acceptance of convolutional neural networks for automatic classification of dopamine transporter SPECT in the diagnosis of clinically uncertain parkinsonian syndromes

Authors: Mahmood Nazari, Andreas Kluge, Ivayla Apostolova, Susanne Klutmann, Sharok Kimiaei, Michael Schroeder, Ralph Buchert

Published in: European Journal of Nuclear Medicine and Molecular Imaging | Issue 4/2022

Abstract

Purpose

Deep convolutional neural networks (CNN) provide high accuracy for automatic classification of dopamine transporter (DAT) SPECT images. However, CNN are inherently black-box in nature lacking any kind of explanation for their decisions. This limits their acceptance for clinical use. This study tested layer-wise relevance propagation (LRP) to explain CNN-based classification of DAT-SPECT in patients with clinically uncertain parkinsonian syndromes.

Methods

The study retrospectively included 1296 clinical DAT-SPECT with visual binary interpretation as “normal” or “reduced” by two experienced readers as standard-of-truth. A custom-made CNN was trained with 1008 randomly selected DAT-SPECT. The remaining 288 DAT-SPECT were used to assess classification performance of the CNN and to test LRP for explanation of the CNN-based classification.

Results

Overall accuracy, sensitivity, and specificity of the CNN were 95.8%, 92.8%, and 98.7%, respectively. LRP provided relevance maps that were easy to interpret in each individual DAT-SPECT. In particular, the putamen in the hemisphere most affected by nigrostriatal degeneration was the most relevant brain region for CNN-based classification in all reduced DAT-SPECT. Some misclassified DAT-SPECT showed an “inconsistent” relevance map more typical for the true class label.

Conclusion

LRP is useful to provide explanation of CNN-based decisions in individual DAT-SPECT and, therefore, can be recommended to support CNN-based classification of DAT-SPECT in clinical routine. Total computation time of 3 s is compatible with busy clinical workflow. The utility of “inconsistent” relevance maps to identify misclassified cases requires further investigation.

Available only for authorised users

Choi H. Deep learning in nuclear medicine and molecular imaging: current perspectives and future directions. Nucl Med Molec Imag. 2018;52:109–18. https://doi.org/10.1007/s13139-017-0504-7.CrossRef

Sakai K, Yamada K. Machine learning studies on major brain diseases: 5-year trends of 2014–2018. Jpn J Radiol. 2019;37:34–72. https://doi.org/10.1007/s11604-018-0794-4.CrossRefPubMed

LeCun Y, Bengio Y, Hinton G. Deep learning. Nature. 2015;521:436–44. https://doi.org/10.1038/nature14539.CrossRef

Litjens G, Kooi T, Bejnordi BE, Setio AAA, Ciompi F, Ghafoorian M, et al. A survey on deep learning in medical image analysis. Med Image Anal. 2017;42:60–88. https://doi.org/10.1016/j.media.2017.07.005.CrossRefPubMed

Castelvecchi D. Can we open the black box of AI? Nature News. 2016;538:20–1.CrossRef

Bach S, Binder A, Montavon G, Klauschen F, Muller KR, Samek W. On pixel-wise explanations for non-linear classifier decisions by layer-wise relevance propagation. Plos One. 2015;10. doi:ARTN e013014010.1371/journal.pone.0130140.

Montavon G, Lapuschkin S, Binder A, Samek W, Muller KR. Explaining nonlinear classification decisions with deep Taylor decomposition. Pattern Recogn. 2017;65:211–22. https://doi.org/10.1016/j.patcog.2016.11.008.CrossRef

Montavon G, Binder A, Lapuschkin S, Samek W, Müller K-R. Layer-wise relevance propagation: an overview. In: Samek W, Montavon G, Vedaldi A, Hansen LK, Müller K-R, editors. Explainable AI: Interpreting. Explaining and Visualizing Deep Learning: Springer; 2019. p. 193–209.CrossRef

Bohle M, Eitel F, Weygandt M, Ritter K. Layer-wise relevance propagation for explaining deep neural network decisions in MRI-based Alzheimer’s disease classification. Front Aging Neurosci. 2019;11:194. https://doi.org/10.3389/fnagi.2019.00194.CrossRefPubMedPubMedCentral

10.

Darcourt J, Booij J, Tatsch K, Varrone A, Vander Borght T, Kapucu OL, et al. EANM procedure guidelines for brain neurotransmission SPECT using (123)I-labelled dopamine transporter ligands, version 2. Eur J Nucl Med Mol Imaging. 2010;37:443–50. https://doi.org/10.1007/s00259-009-1267-x.CrossRefPubMed

11.

Sjoholm H, Bratlid T, Sundsfjord J. I-123-beta-CIT SPECT demonstrates increased presynaptic dopamine transporter binding sites in basal ganglia in vivo in schizophrenia. Psychopharmacology. 2004;173:27–31. https://doi.org/10.1007/s00213-003-1700-y.CrossRefPubMed

12.

Tossici-Bolt L, Dickson JC, Sera T, Booij J, Asenbaun-Nan S, Bagnara MC, et al. [I-123] FP-CIT ENC-DAT normal database: the impact of the reconstruction and quantification methods. Ejnmmi Phys. 2017;4. doi:https://doi.org/10.1186/s40658-017-0175-6.

13.

Acton PD, Friston KJ. Statistical parametric mapping in functional neuroimaging: beyond PET and fMRI activation studies. Eur J Nucl Med. 1998;25:663–7.PubMed

14.

Kupitz D, Apostolova I, Lange C, Ulrich G, Amthauer H, Brenner W, et al. Global scaling for semi-quantitative analysis in FP-CIT SPECT. Nuklearmed-Nucl Med. 2014;53:234–41. https://doi.org/10.3413/Nukmed-0659-14-04.CrossRef

15.

Koch W, Unterrainer M, Xiong G, Bartenstein P, Diemling M, Varrone A, et al. Extrastriatal binding of [(1)(2)(3)I]FP-CIT in the thalamus and pons: gender and age dependencies assessed in a European multicentre database of healthy controls. Eur J Nucl Med Mol Imaging. 2014;41:1938–46. https://doi.org/10.1007/s00259-014-2785-8.CrossRefPubMed

16.

Apostolova I, Taleb DS, Lipp A, Galazky I, Kupitz D, Lange C, et al. Utility of follow-up dopamine transporter SPECT with 123I-FP-CIT in the diagnostic workup of patients with clinically uncertain Parkinsonian syndrome. Clin Nucl Med. 2017;42:589–94. https://doi.org/10.1097/RLU.0000000000001696.CrossRefPubMed

17.

Buchert R, Buhmann C, Apostolova I, Meyer PT, Gallinat J. Nuclear imaging in the diagnosis of clinically uncertain Parkinsonian syndromes. Dtsch Arztebl Int. 2019;116:747–54. https://doi.org/10.3238/arztebl.2019.0747.CrossRefPubMedPubMedCentral

18.

Wenzel M, Milletari F, Kruger J, Lange C, Schenk M, Apostolova I, et al. Automatic classification of dopamine transporter SPECT: deep convolutional neural networks can be trained to be robust with respect to variable image characteristics. Eur J Nucl Med Mol Imaging. 2019;46:2800–11. https://doi.org/10.1007/s00259-019-04502-5.CrossRefPubMed

19.

Sabour S, Frosst N, Hinton GE. Dynamic routing between capsules. arXiv. 2017:arXiv:1710.09829.

20.

Dobbin KK, Simon RM. Optimally splitting cases for training and testing high dimensional classifiers. BMC Med Genomics. 2011;4:31. https://doi.org/10.1186/1755-8794-4-31.CrossRefPubMedPubMedCentral

21.

Hajian-Tilaki K. Sample size estimation in diagnostic test studies of biomedical informatics. J Biomed Inform. 2014;48:193–204. https://doi.org/10.1016/j.jbi.2014.02.013.CrossRefPubMed

22.

Samek W, Müller K-R. Towards explainable artificial intelligence. In: Samek W, Montavon G, Vedaldi A, Hansen LK, Müller K-R, editors. Explainable AI: interpreting, explaining and visualizing deep learning. Cham: Springer Nature; 2019. pp. 5–22.

23.

Kohlbrenner M, Bauer A, Nakajima S, Binder A, Samek W, Lapuschkin S. Towards best practice in explaining neural network decisions with LRP. In: Proceedings of the 2020 International Joint Conference on Neural Networks. Red Hook, NY: Curran Associates; 2020. pp. 1–7.

24.

The shattered gradients problem: If resnets are the answer, then what is the question? In: Precup D, Teh YW, editors. Proceedings of the 34^th International Conference on Machine Learning. Sydney: PMLR; 2017. pp. 342–50.

25.

Bach S, Binder A, Müller K-R, Samek W. Controlling explanatory heatmap resolution and semantics via decomposition depth. In: Proceedings of the 2016 IEEE International Conference on Image Processing. Red Hook, NY: Curran Associates; 2016. pp. 2271–5.

26.

Shrikumar A, Greenside P, Kundaje A. Learning important features through propagating activation differences. In: Precup D, Teh YW, editors. Proceedings of the 34^th International Conference on Machine Learning. Sydney: PMLR; 2017. pp. 3145–53.

27.

Petsiuk V, Das A, Saenko K. Rise: randomized input sampling for explanation of black-box models. arXiv preprint 2018; arXiv180607421.

28.

Lundberg SM, Lee S-I. A unifed approach to interpreting model predictions. In: von Luxburg U, Guyon I, Bengio S, Wallach H, Fergus R, editors. In: Proceedings of the 31^st International Conference on Neural Information Processing Systems. Red Hook, NY: Curran Associates; 2017. pp. 4768–77.

29.

Ribeiro MT, Singh S, Guestrin C. “Why should I trust you?”: explaining the predictions of any classifier. In: Proceedings of the 22^nd ACM SIGKDD international conference on knowledge discovery and data mining. New York, NY: Association for Computing Machinery; 2016. pp. 1135–44.

30.

Eitel F, Soehler E, Bellmann-Strobl J, Brandt AU, Ruprecht K, Giess RM, et al. Uncovering convolutional neural network decisions for diagnosing multiple sclerosis on conventional MRI using layer-wise relevance propagation. Neuroimage-Clin. 2019;24. doi:ARTN 10200310.1016/j.nicl.2019.102003.

31.

Morbelli S, Esposito G, Arbizu J, Barthel H, Boellaard R, Bohnen NI, et al. EANM practice guideline/SNMMI procedure standard for dopaminergic imaging in Parkinsonian syndromes 1.0. Eur J Nucl Med Mol. 2020;I(47):1885–912. https://doi.org/10.1007/s00259-020-04817-8.CrossRef

32.

Booij J, Speelman JD, Horstink MW, Wolters EC. The clinical benefit of imaging striatal dopamine transporters with [123I]FP-CIT SPET in differentiating patients with presynaptic parkinsonism from those with other forms of parkinsonism. Eur J Nucl Med. 2001;28:266–72.CrossRef

33.

Dotinga M, van Dijk JD, Vendel BN, Slump CH, Portman AT, van Dalen JA. Clinical value of machine learning-based interpretation of I-123 FP-CIT scans to detect Parkinson’s disease: a two-center study. Ann Nucl Med. 2021;35:378–85. https://doi.org/10.1007/s12149-021-01576-w.CrossRefPubMed

34.

Castillo-Barnes D, Martinez-Murcia FJ, Ortiz A, Salas-Gonzalez D, RamIrez J, Gorriz JM. Morphological characterization of functional brain imaging by isosurface analysis in Parkinson’s disease. International Journal of Neural Systems. 2020;30. doi:Artn 205004410.1142/S0129065720500446.

35.

Segovia F, Gorriz JM, Ramirez J, Martinez-Murcia FJ, Castillo-Barnes D. Assisted diagnosis of Parkinsonism based on the striatal morphology. International Journal of Neural Systems. 2019;29. doi:Artn 195001110.1142/S0129065719500114.

36.

Nicastro N, Wegrzyk J, Preti MG, Fleury V, Van de Ville D, Garibotto V, et al. Classification of degenerative parkinsonism subtypes by support-vector-machine analysis and striatal I-123-FP-CIT indices. J Neurol. 2019;266:1771–81. https://doi.org/10.1007/s00415-019-09330-z.CrossRefPubMedPubMedCentral

37.

Hsu SY, Lin HC, Chen TB, Du WC, Hsu YH, Wu YC, et al. Feasible classified models for Parkinson disease from Tc-99m-TRODAT-1 SPECT imaging. Sensors-Basel. 2019;19. doi:ARTN 174010.3390/s19071740.

38.

Iwabuchi Y, Nakahara T, Kameyama M, Yamada Y, Hashimoto M, Matsusaka Y, et al. Impact of a combination of quantitative indices representing uptake intensity, shape, and asymmetry in DAT SPECT using machine learning: comparison of different volume of interest settings. Ejnmmi Res. 2019;9. doi:ARTN 710.1186/s13550–019–0477-x.

39.

Castillo-Barnes D, Ramirez J, Segovia F, Martinez-Murcia FJ, Saias-Gonzalez D, Gorriz JM. Robust ensemble classification methodology for I123-Ioflupane SPECT images and multiple heterogeneous biomarkers in the diagnosis of Parkinson’s disease. Front Neuroinform. 2018;12. doi:ARTN 5310.3389/fninf.2018.00053.

40.

Oliveira FPM, Faria DB, Costa DC, Castelo-Branco M, Tavares J. Extraction, selection and comparison of features for an effective automated computer-aided diagnosis of Parkinson’s disease based on [(123)I]FP-CIT SPECT images. Eur J Nucl Med Mol Imaging. 2018;45:1052–62. https://doi.org/10.1007/s00259-017-3918-7.CrossRefPubMed

41.

Taylor JC, Fenner JW. Comparison of machine learning and semi-quantification algorithms for (I123)FP-CIT classification: the beginning of the end for semi-quantification? Ejnmmi Phys. 2017;4:1-20. doi:ARTN 2910.1186/s40658-017-0196-1.

42.

Palumbo B, Fravolini ML, Buresta T, Pompili F, Forini N, Nigro P, et al. Diagnostic accuracy of Parkinson disease by support vector machine (SVM) analysis of I-123-FP-CIT brain SPECT data. Medicine. 2014;93. doi:ARTN e22810.1097/MD.0000000000000228.

43.

Huertas-Fernandez I, Garcia-Gomez FJ, Garcia-Solis D, Benitez-Rivero S, Marin-Oyaga VA, Jesus S, et al. Machine learning models for the differential diagnosis of vascular parkinsonism and Parkinson’s disease using [I-123]FP-CIT SPECT. Eur J Nucl Med Mol. 2015;I(42):112–9. https://doi.org/10.1007/s00259-014-2882-8.CrossRef

44.

Iwabuchi Y, Kameyama M, Matsusaka Y, Narimatsu H, Hashimoto M, Seki M, et al. A diagnostic strategy for Parkinsonian syndromes using quantitative indices of DAT SPECT and MIBG scintigraphy: an investigation using the classification and regression tree analysis. Eur J Nucl Med Mol Imaging. 2021. https://doi.org/10.1007/s00259-020-05168-0.CrossRefPubMedPubMedCentral

45.

Cascianelli S, Tranfaglia C, Fravolini ML, Bianconi F, Minestrini M, Nuvoli S, et al. Right putamen and age are the most discriminant features to diagnose Parkinson’s disease by using (123)I-FP-CIT brain SPET data by using an artificial neural network classifier, a classification tree (ClT). Hell J Nucl Med. 2017;20(Suppl):165.PubMed

46.

Salmanpour MR, Shamsaei M, Saberi A, Hajianfar G, Soltanian-Zadeh H, Rahmim A. Robust identification of Parkinson’s disease subtypes using radiomics and hybrid machine learning. Comput Biol Med. 2021;129. doi:ARTN 10414210.1016/j.compbiomed.2020.104142.

47.

Chien CY, Hsu SW, Lee TL, Sung PS, Lin CC. Using artificial neural network to discriminate Parkinson’s disease from other Parkinsonisms by focusing on putamen of dopamine transporter SPECT images. Biomedicines. 2021;9. doi:ARTN 1210.3390/biomedicines9010012.

48.

Magesh PR, Myloth RD, Tom RJ. An explainable machine learning model for early detection of Parkinson’s disease using LIME on DaTSCAN imagery. Comput Biol Med. 2020;126. doi:ARTN 10404110.1016/j.compbiomed.2020.104041.

49.

Ozsahin I, Sekeroglu B, Pwavodi PC, Mok GSP. High-accuracy automated diagnosis of Parkinson’s disease. Curr Med Imaging. 2020;16:688–94. https://doi.org/10.2174/1573405615666190620113607.CrossRefPubMed

50.

Ortiz A, Munilla J, Martinez-Ibanez M, Gorriz JM, Ramirez J, Salas-Gonzalez D. Parkinson’s disease detection using isosurfaces-based features and convolutional neural networks. Front Neuroinform. 2019;13. doi:ARTN 4810.3389/fninf.2019.00048.

51.

Martinez-Murcia FJ, Gorriz JM, Ramirez J, Ortiz A. Convolutional neural networks for neuroimaging in Parkinson’s disease: is preprocessing needed? Int J Neural Syst. 2018:1850035. https://doi.org/10.1142/S0129065718500351.

52.

Kim DH, Wit H, Thurston M. Artificial intelligence in the diagnosis of Parkinson’s disease from ioflupane-123 single-photon emission computed tomography dopamine transporter scans using transfer learning. Nucl Med Commun. 2018;39:887–93. https://doi.org/10.1097/MNM.0000000000000890.CrossRefPubMed

53.

Choi H, Ha S, Im HJ, Paek SH, Lee DS. Refining diagnosis of Parkinson’s disease with deep learning-based interpretation of dopamine transporter imaging. Neuroimage Clin. 2017;16:586–94. https://doi.org/10.1016/j.nicl.2017.09.010.CrossRefPubMedPubMedCentral

54.

Zhang YC, Kagen AC. Machine learning interface for medical image analysis. J Digit Imaging. 2017;30:615–21. https://doi.org/10.1007/s10278-016-9910-0.CrossRefPubMed

55.

Palumbo B, Fravolini ML, Nuvoli S, Spanu A, Paulus KS, Schillaci O, et al. Comparison of two neural network classifiers in the differential diagnosis of essential tremor and Parkinson’s disease by I-123-FP-CIT brain SPECT. Eur J Nucl Med Mol. 2010;I(37):2146–53. https://doi.org/10.1007/s00259-010-1481-6.CrossRef

56.

Acton PD, Newberg A. Artificial neural network classifier for the diagnosis of Parkinson’s disease using [Tc-99m] TRODAT-1 and SPECT. Phys Med Biol. 2006;51:3057–66. https://doi.org/10.1088/0031-9155/51/12/004.CrossRefPubMed

57.

Mohammed F, He XJ, Lin YG. An easy-to-use deep-learning model for highly accurate diagnosis of Parkinson’s disease using SPECT images. Comput Med Imag Grap. 2021;87. doi:ARTN 10181010.1016/j.compmedimag.2020.101810.

58.

Huang GH, Lin CH, Cai YR, Chen TB, Hsu SY, Lu NH, et al. Multiclass machine learning classification of functional brain images for Parkinson’s disease stage prediction. Stat Anal Data Min. 2020;13:508–23. https://doi.org/10.1002/sam.11480.CrossRef

59.

Booij J, van de Giessen E, Hesse S, SabriO.Comments on Eusebio, , et al. Voxel-based analysis of whole-brain effects of age and gender on dopamine transporter SPECT imaging in healthy subjects. Eur J Nucl Med Mol. 2013;I(40):143–4. https://doi.org/10.1007/s00259-012-2267-9.CrossRef

60.

Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition. arXiv:14091556. 2014.

Title: Explainable AI to improve acceptance of convolutional neural networks for automatic classification of dopamine transporter SPECT in the diagnosis of clinically uncertain parkinsonian syndromes
Authors: Mahmood Nazari
Andreas Kluge
Ivayla Apostolova
Susanne Klutmann
Sharok Kimiaei
Michael Schroeder
Ralph Buchert
Publication date: 01-03-2022
Publisher: Springer Berlin Heidelberg
Keywords: Artificial Intelligence
Parkinson's Disease
Published in: European Journal of Nuclear Medicine and Molecular Imaging / Issue 4/2022
Print ISSN: 1619-7070
Electronic ISSN: 1619-7089
DOI: https://doi.org/10.1007/s00259-021-05569-9

Keynote webinar | Spotlight on medication adherence

Springer Medicine

Explainable AI to improve acceptance of convolutional neural networks for automatic classification of dopamine transporter SPECT in the diagnosis of clinically uncertain parkinsonian syndromes

Abstract

Purpose

Methods

Results

Conclusion

Keynote webinar | Spotlight on medication adherence

Springer Medicine

Abstract

Purpose

Methods

Results

Conclusion

Please log in to get access to this content

Other articles of this Issue 4/2022

Joint EANM/SNMMI/ESTRO practice recommendations for the use of 2-[18F]FDG PET/CT external beam radiation treatment planning in lung cancer V1.0

Added value of semiquantitative analysis of brain FDG-PET for the differentiation between MCI-Lewy bodies and MCI due to Alzheimer’s disease

EANM position on the in-house preparation of radiopharmaceuticals

Prediction of post-stroke cognitive impairment using brain FDG PET: deep learning-based approach

A sigh of relief: vaccine-associated hypermetabolic lymphadenopathy following the third COVID-19 vaccine dose is short in duration and uncommonly interferes with the interpretation of [18F]FDG PET-CT studies performed in oncologic patients

PET/CT imaging of head-and-neck and pancreatic cancer in humans by targeting the “Cancer Integrin” αvβ6 with Ga-68-Trivehexin