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

다국어 초록 (Multilingual 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.

참고문헌 (Reference)

1 Hayek CS, "Wavelet processing of systolic murmurs to assist with clinical diagnosis of heart disease" 37 : 263-270, 2003

2 McCarley P, "Vascular access blood flow monitoring reduces access morbidity and costs" 60 : 1164-1172, 2001

3 Lin YP, "Spiral computed tomographic angiography--a new technique for evaluation of vascular access in hemodialysis patients" 18 : 117-122, 1998

4 Sacks D, "Society of interventional radiology clinical practice guidelines" 14 (14): S199-S202, 2003

5 Chawla NV, "SMOTE : synthetic minority over-sampling technique" 16 : 321-357, 2002

6 Palanisamy K, "Rethinking CNN models for audio classification"

7 Vasudevan RS, "Persistent value of the stethoscope in the age of COVID-19" 133 : 1143-1150, 2020

8 Wang HY, "Novel noninvasive approach for detecting arteriovenous fistula stenosis" 61 : 1851-1857, 2014

9 Messner E, "Multi-channel lung sound classification with convolutional recurrent neural networks" 122 : 103831-, 2020

10 Bardou D, "Lung sounds classification using convolutional neural networks" 88 : 58-69, 2018

11 McFee B, "Librosa: audio and music signal analysis in python" SciPy 18-25, 2015

12 Lok CE, "KDOQI clinical practice guideline for vascular access : 2019update" 75 (75): S1-S164, 2020

13 Akay M, "Investigating the effects of vasodilator drugs on the turbulent sound caused by femoral artery stenosis using short-term Fourier and wavelet transform methods" 41 : 921-928, 1994

14 Selvaraju RR, "Grad-cam : visual explanations from deep networks via gradient-based localization" IEEE 618-626, 2017

15 Sato T, "Evaluation of blood access dysfunction based on a wavelet transform analysis of shunt murmurs" 9 : 97-104, 2006

16 Nanni L, "Ensemble of convolutional neural networks to improve animal audio classification" 2020 : 8-, 2020

17 Tan M, "Efficientnet : rethinking model scaling for convolutional neural networks" PMLR 6105-6114, 2019

18 Huang G, "Densely connected convolutional networks" IEEE 4700-4708, 2017

19 He K, "Deep residual learning for image recognition" IEEE 770-778, 2016

20 Glangetas A, "Deep learning diagnostic and riskstratification pattern detection for COVID-19 in digital lung auscultations : clinical protocol for a case-control and prospective cohort study" 21 : 103-, 2021

21 Mansy HA, "Computerised analysis of auscultatory sounds associated with vascular patency of haemodialysis access" 43 : 56-62, 2005

22 Brescia MJ, "Chronic hemodialysis using venipuncture and a surgically created arteriovenous fistula" 275 : 1089-1092, 1966

23 Tessitore N, "Can blood flow surveillance and pre-emptive repair of subclinical stenosis prolong the useful life of arteriovenous fistulae? A randomized controlled study" 19 : 2325-2333, 2004

24 Kingma DP, "Adam: a method for stochastic optimization"

25 Bountouris I, "A review of percutaneous transluminal angioplasty in hemodialysis fistula" 2018 : 1420136-, 2018

26 Sehgal A, "A convolutional neural network smartphone app for real-time voice activity detection" 6 : 9017-9026, 2018