Korean Journal of Radiology

The Korean Society of Radiology (대한영상의학회)
권호 표지
DOI QR코드

DOI QR Code

Feasibility of Deep Learning-Based Analysis of Auscultation for Screening Significant Stenosis of Native Arteriovenous Fistula for Hemodialysis Requiring Angioplasty

  • Jae Hyon Park (Department of Radiology, Yonsei University College of Medicine) ;
  • Insun Park (Department of Surgery, Yonsei University College of Medicine) ;
  • Kichang Han (Department of Radiology, Yonsei University College of Medicine) ;
  • Jongjin Yoon (Department of Radiology, Yonsei University College of Medicine) ;
  • Yongsik Sim (Department of Radiology, Yonsei University College of Medicine) ;
  • Soo Jin Kim (Department of Anesthesiology and Pain Medicine, Seoul National University Bundang Hospital) ;
  • Jong Yun Won (Department of Radiology, Yonsei University College of Medicine) ;
  • Shina Lee (Department of Internal Medicine, College of Medicine, Ewha Womans University) ;
  • Joon Ho Kwon (Department of Radiology, Yonsei University College of Medicine) ;
  • Sungmo Moon (Department of Radiology, Yonsei University College of Medicine) ;
  • Gyoung Min Kim (Department of Radiology, Yonsei University College of Medicine) ;
  • Man-deuk Kim (Department of Radiology, Yonsei University College of Medicine)
  • Received : 2022.06.04
  • Accepted : 2022.08.05
  • Published : 2022.10.01
Download PDF
⟨ Previous Next ⟩

Abstract

Objective: To investigate the feasibility of using a deep learning-based analysis of auscultation data to predict significant stenosis of arteriovenous fistulas (AVF) in patients undergoing hemodialysis requiring percutaneous transluminal angioplasty (PTA). Materials and Methods: Forty patients (24 male and 16 female; median age, 62.5 years) with dysfunctional native AVF were prospectively recruited. Digital sounds from the AVF shunt were recorded using a wireless electronic stethoscope before (pre-PTA) and after PTA (post-PTA), and the audio files were subsequently converted to mel spectrograms, which were used to construct various deep convolutional neural network (DCNN) models (DenseNet201, EfficientNetB5, and ResNet50). The performance of these models for diagnosing ≥ 50% AVF stenosis was assessed and compared. The ground truth for the presence of ≥ 50% AVF stenosis was obtained using digital subtraction angiography. Gradient-weighted class activation mapping (Grad-CAM) was used to produce visual explanations for DCNN model decisions. Results: Eighty audio files were obtained from the 40 recruited patients and pooled for the study. Mel spectrograms of "pre-PTA" shunt sounds showed patterns corresponding to abnormal high-pitched bruits with systolic accentuation observed in patients with stenotic AVF. The ResNet50 and EfficientNetB5 models yielded an area under the receiver operating characteristic curve of 0.99 and 0.98, respectively, at optimized epochs for predicting ≥ 50% AVF stenosis. However, Grad-CAM heatmaps revealed that only ResNet50 highlighted areas relevant to AVF stenosis in the mel spectrogram. Conclusion: Mel spectrogram-based DCNN models, particularly ResNet50, successfully predicted the presence of significant AVF stenosis requiring PTA in this feasibility study and may potentially be used in AVF surveillance.

Keywords

References

  1. Brescia MJ, Cimino JE, Appel K, Hurwich BJ. Chronic hemodialysis using venipuncture and a surgically created arteriovenous fistula. N Engl J Med 1966;275:1089-1092
  2. Lok CE, Huber TS, Lee T, Shenoy S, Yevzlin AS, Abreo K, et al. KDOQI clinical practice guideline for vascular access: 2019 update. Am J Kidney Dis 2020;75(4 Suppl 2):S1-S164
  3. Lin YP, Wu MH, Ng YY, Lee RC, Liou JK, Yang WC, et al. Spiral computed tomographic angiography--a new technique for evaluation of vascular access in hemodialysis patients. Am J Nephrol 1998;18:117-122
  4. Bountouris I, Kritikou G, Degermetzoglou N, Avgerinos KI. A review of percutaneous transluminal angioplasty in hemodialysis fistula. Int J Vasc Med 2018;2018:1420136
  5. Bardou D, Zhang K, Ahmad SM. Lung sounds classification using convolutional neural networks. Artif Intell Med 2018;88:58-69
  6. Messner E, Fediuk M, Swatek P, Scheidl S, Smolle-Juttner FM, Olschewski H, et al. Multi-channel lung sound classification with convolutional recurrent neural networks. Comput Biol Med 2020;122:103831
  7. Sacks D, McClenny TE, Cardella JF, Lewis CA. Society of interventional radiology clinical practice guidelines. J Vasc Interv Radiol 2003;14(9 Pt 2):S199-S202
  8. McFee B, Raffel C, Liang D, Ellis DP, McVicar M, Battenberg E, et al. Librosa: audio and music signal analysis in python. Proceedings of the 14th Python in Science Conference; 2015 Jul 6-12; Austin, TX, USA: SciPy; 2015. p.18-25
  9. Palanisamy K, Singhania D, Yao A. Rethinking CNN models for audio classification. arXiv [Preprint]. 2020 [cited 2020 Jul 1]. Available at: https://doi.org/10.48550/arXiv.2007.11154
  10. Sehgal A, Kehtarnavaz N. A convolutional neural network smartphone app for real-time voice activity detection. IEEE Access 2018;6:9017-9026
  11. Chawla NV, Bowyer KW, Hall LO, Kegelmeyer WP. SMOTE: synthetic minority over-sampling technique. J Artif Intell Res 2002;16:321-357
  12. Huang G, Liu Z, Van Der Maaten L, Weinberger KQ. Densely connected convolutional networks. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR); 2017 Jul 22-25; Honolulu, HI, USA: IEEE; 2017. p.4700-4708
  13. Tan M, Le Q. Efficientnet: rethinking model scaling for convolutional neural networks. Proceedings of the 36th International Conference on Machine Learning; 2019 Jun 9-15; Long Beach, CA, USA: PMLR; 2019. p.6105-6114
  14. He K, Zhang X, Ren S, Sun J. Deep residual learning for image recognition. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR); 2016 Jun 27-30; Las Vegas, NV, USA: IEEE; 2016. p.770-778
  15. Kingma DP, Ba J. Adam: a method for stochastic optimization. arXiv [Preprint]. 2014 [cited 2020 Jul 1]. Available at: https://doi.org/10.48550/arXiv.1412.6980
  16. Selvaraju RR, Cogswell M, Das A, Vedantam R, Parikh D, Batra D. Grad-cam: visual explanations from deep networks via gradient-based localization. Proceedings of the IEEE International Conference on Computer Vision (ICCV); 2017 Jul 22-25; Venice, Italy: IEEE; 2017. p.618-626
  17. Wang HY, Wu CH, Chen CY, Lin BS. Novel noninvasive approach for detecting arteriovenous fistula stenosis. IEEE Trans Biomed Eng 2014;61:1851-1857
  18. Hayek CS, Thompson WR, Tuchinda C, Wojcik RA, Lombardo JS. Wavelet processing of systolic murmurs to assist with clinical diagnosis of heart disease. Biomed Instrum Technol 2003;37:263-270
  19. Akay M, Akay YM, Welkowitz W, Lewkowicz S. Investigating the effects of vasodilator drugs on the turbulent sound caused by femoral artery stenosis using short-term Fourier and wavelet transform methods. IEEE Trans Biomed Eng 1994;41:921-928
  20. Mansy HA, Hoxie SJ, Patel NH, Sandler RH. Computerised analysis of auscultatory sounds associated with vascular patency of haemodialysis access. Med Biol Eng Comput 2005;43:56-62
  21. Sato T, Tsuji K, Kawashima N, Agishi T, Toma H. Evaluation of blood access dysfunction based on a wavelet transform analysis of shunt murmurs. J Artif Organs 2006;9:97-104
  22. Glangetas A, Hartley MA, Cantais A, Courvoisier DS, Rivollet D, Shama DM, et al. Deep learning diagnostic and risk-stratification pattern detection for COVID-19 in digital lung auscultations: clinical protocol for a case-control and prospective cohort study. BMC Pulm Med 2021;21:103
  23. Vasudevan RS, Horiuchi Y, Torriani FJ, Cotter B, Maisel SM, Dadwal SS, et al. Persistent value of the stethoscope in the age of COVID-19. Am J Med 2020;133:1143-1150
  24. McCarley P, Wingard RL, Shyr Y, Pettus W, Hakim RM, Ikizler TA. Vascular access blood flow monitoring reduces access morbidity and costs. Kidney Int 2001;60:1164-1172
  25. Tessitore N, Lipari G, Poli A, Bedogna V, Baggio E, Loschiavo C, et al. Can blood flow surveillance and pre-emptive repair of subclinical stenosis prolong the useful life of arteriovenous fistulae? A randomized controlled study. Nephrol Dial Transplant 2004;19:2325-2333
  26. Nanni L, Costa YM, Aguiar RL, Mangolin RB, Brahnam S, Silla CN. Ensemble of convolutional neural networks to improve animal audio classification. J Audio Speech Music Proc 2020;2020:8