Breathe out the Secret of the Lung: Video Classification of Exhaled Flows from Normal and Asthmatic Lung Models Using CNN-Long Short-Term Memory Networks DOI Creative Commons
H. Talaat, Xiuhua Si, Jinxiang Xi

et al.

Journal of Respiration, Journal Year: 2023, Volume and Issue: 3(4), P. 237 - 257

Published: Dec. 14, 2023

In this study, we present a novel approach to differentiate normal and diseased lungs based on exhaled flows from 3D-printed lung models simulating asthmatic conditions. By leveraging the sequential learning capacity of Long Short-Term Memory (LSTM) network automatic feature extraction convolutional neural networks (CNN), evaluated feasibility detection staging airway constrictions. Two (D1, D2) with increasing levels severity were generated by decreasing bronchiolar calibers in right upper lobe (D0). Expiratory recorded mid-sagittal plane using high-speed camera at 1500 fps. addition baseline flow rate (20 L/min) which trained verified, two additional rates (15 L/min 10 considered evaluate network’s robustness deviations. Distinct patterns vortex dynamics observed among three disease states (D0, D1, across rates. The AlexNet-LSTM proved be robust, maintaining perfect performance three-class classification when deviated recommendation 25%, still performed reasonably (72.8% accuracy) despite 50% deviation. GoogleNet-LSTM also showed satisfactory (91.5% 25% deviation but exhibited low (57.7% was 50%. Considering effects task, video classifications only slightly outperformed those images (i.e., 3–6%). occlusion sensitivity analyses distinct heat maps specific state.

Language: Английский

Schlieren imaging and video classification of alphabet pronunciations: exploiting phonetic flows for speech recognition and speech therapy DOI Creative Commons
H. Talaat, Kian Barari, Xiuhua April

et al.

Visual Computing for Industry Biomedicine and Art, Journal Year: 2024, Volume and Issue: 7(1)

Published: May 22, 2024

Abstract Speech is a highly coordinated process that requires precise control over vocal tract morphology/motion to produce intelligible sounds while simultaneously generating unique exhaled flow patterns. The schlieren imaging technique visualizes airflows with subtle density variations. It hypothesized speech flows captured by schlieren, when analyzed using hybrid of convolutional neural network (CNN) and long short-term memory (LSTM) network, can recognize alphabet pronunciations, thus facilitating automatic recognition disorder therapy. This study evaluates the feasibility CNN-based video classification differentiate corresponding first four alphabets: /A/, /B/, /C/, /D/. A optical system was developed, pronunciations were recorded for two participants at an acquisition rate 60 frames per second. total 640 clips, each lasting 1 s, utilized train test CNN-LSTM network. Acoustic analyses conducted understand phonetic differences among alphabets. trained separately on datasets varying sizes (i.e., 20, 30, 40, 50 videos alphabet), all achieving 95% accuracy in classifying same participant. However, network’s performance declined tested from different participant, dropping around 44%, indicating significant inter-participant variability pronunciation. Retraining both improved 93% second Analysis misclassified indicated factors such as low quality disproportional head size affected accuracy. These results highlight potential CNN-assisted therapy articulation flows, although challenges remain expanding set participant cohort.

Language: Английский

Citations

4
Developing Approaches to Incorporate Donor Lung CT Images into Machine Learning Models to Predict Severe Primary Graft Dysfunction after Lung Transplantation DOI Creative Commons
W. F. Mader, Inez Y. Oh, Yixuan Luo

et al.

American Journal of Transplantation, Journal Year: 2025, Volume and Issue: unknown

Published: Feb. 1, 2025

Primary graft dysfunction (PGD) is a common complication after lung transplantation associated with poor outcomes. Although risk factors have been identified, the complex interactions between clinical variables affecting PGD are not well understood, which can complicate decisions about donor acceptance. Previously, we developed machine learning (ML) model to predict grade 3 using and recipient electronic health record (EHR) data, but it lacked granular information from CT scans, routinely assessed during offer review. In this study, used gated approach determine optimal methods for analyzing scans among patients receiving first-time, bilateral transplants at single center over 10 years. We four computer vision approaches fused best EHR data three points in ML process. A total of 160 had donor-lung analysis. The imaging-only employed 3D ResNet model, yielding median (IQR) AUROC AUPRC 0.63 (0.49 - 0.72) 0.48 (0.35 0.6), respectively. Combining imaging late fusion provided highest performance, 0.74 (0.59 0.85) 0.61 (0.47 0.72),

Language: Английский

Citations

0
Concatenated CNN-Based Pneumonia Detection Using a Fuzzy-Enhanced Dataset DOI Creative Commons
Abror Shavkatovich Buriboev, Dilnoz Muhamediyeva, Holida Primova

et al.

Sensors, Journal Year: 2024, Volume and Issue: 24(20), P. 6750 - 6750

Published: Oct. 21, 2024

Pneumonia is a form of acute respiratory infection affecting the lungs. Symptoms viral and bacterial pneumonia are similar. Rapid diagnosis disease difficult, since polymerase chain reaction-based methods, which have greatest reliability, provide results in few hours, while ensuring high requirements for compliance with analysis technology professionalism personnel. This study proposed Concatenated CNN model detection combined fuzzy logic-based image improvement method. The enhancement process based on new fuzzification refinement algorithm, significantly improved quality feature extraction CCNN model. Four datasets, original upgraded images utilizing entropy, standard deviation, histogram equalization, were utilized to train algorithm. CCNN's performance was demonstrated be by entropy-added dataset producing best results. suggested attained remarkable classification metrics, including 98.9% accuracy, 99.3% precision, 99.8% F1-score, 99.6% recall. Experimental comparisons showed that worked better than traditional resulting higher diagnostic precision. demonstrates how well deep learning models sophisticated techniques work together analyze medical images.

Language: Английский

Citations

2
Data-Driven Discovery of Anomaly-Sensitive Parameters from Uvula Wake Flows Using Wavelet Analyses and Poincaré Maps DOI Creative Commons
Xiuhua Si, Junshi Wang, Haibo Dong

et al.

Acoustics, Journal Year: 2023, Volume and Issue: 5(4), P. 1046 - 1065

Published: Nov. 2, 2023

This study presents a data-driven approach to identifying anomaly-sensitive parameters through multiscale, multifaceted analysis of simulated respiratory flows. The anomalies under consideration include pharyngeal model with three levels constriction (M1, M2, M3) and flapping uvula two types kinematics (K1, K2). Direct numerical simulations (DNS) were implemented solve the wake flows induced by uvula; instantaneous vortex images, as well pressures velocities at seven probes, recorded for twelve cycles. Principal component (PCA), wavelet-based multifractal spectrum scalogram, Poincaré mapping identify parameters. PCA results demonstrated reasonable periodicity images in leading vector space revealed distinct patterns between models varying At higher ranks, gradually decays, eventually transitioning random pattern. spectra scalograms pharynx (P6, P7) show high sensitivity kinematics, pitching mode (K2) having wider left-skewed peak than heaving (K1). Conversely, maps (Vel6, Vel7, P6, exhibit (M1–M3), but not kinematics. parameter anomaly also differs probe site; thus, synergizing measurements from multiple probes properly extracted holds potential localize source snoring estimate collapsibility pharynx.

Language: Английский

Citations

5
Breathe out the Secret of the Lung: Video Classification of Exhaled Flows from Normal and Asthmatic Lung Models Using CNN-Long Short-Term Memory Networks DOI Creative Commons
H. Talaat, Xiuhua Si, Jinxiang Xi

et al.

Journal of Respiration, Journal Year: 2023, Volume and Issue: 3(4), P. 237 - 257

Published: Dec. 14, 2023

In this study, we present a novel approach to differentiate normal and diseased lungs based on exhaled flows from 3D-printed lung models simulating asthmatic conditions. By leveraging the sequential learning capacity of Long Short-Term Memory (LSTM) network automatic feature extraction convolutional neural networks (CNN), evaluated feasibility detection staging airway constrictions. Two (D1, D2) with increasing levels severity were generated by decreasing bronchiolar calibers in right upper lobe (D0). Expiratory recorded mid-sagittal plane using high-speed camera at 1500 fps. addition baseline flow rate (20 L/min) which trained verified, two additional rates (15 L/min 10 considered evaluate network’s robustness deviations. Distinct patterns vortex dynamics observed among three disease states (D0, D1, across rates. The AlexNet-LSTM proved be robust, maintaining perfect performance three-class classification when deviated recommendation 25%, still performed reasonably (72.8% accuracy) despite 50% deviation. GoogleNet-LSTM also showed satisfactory (91.5% 25% deviation but exhibited low (57.7% was 50%. Considering effects task, video classifications only slightly outperformed those images (i.e., 3–6%). occlusion sensitivity analyses distinct heat maps specific state.

Language: Английский

Citations

3