Since the COVID-19 pandemic, interest in non-contact diagnostic technologies has grown, leading to increased research into remote biosignal monitoring. The respiratory rate, widely used in previous studies, offers limited insight into pulmonary volume. To redress this, we propose a thermal imaging-based framework for respiratory segmentation aimed at estimating non-invasive pulmonary function. The proposed method uses an optical flow magnitude-based thresholding technique to automatically extract exhalation frames and segment them into frame sequences. A TransUNet-based network, combining a Convolutional Neural Network (CNN) encoder–decoder architecture with a Transformer module in the bottleneck, is trained on these sequences. The model’s Accuracy, Precision, Recall, IoU, Dice, and F1-score were 0.9832, 0.9833, 0.9830, 0.9651, 0.9822, and 0.9831, respectively, which results demonstrate high segmentation performance. The method enables the respiratory volume to be estimated by detecting exhalation behavior, suggesting its potential as a non-contact tool to monitor pulmonary function and estimate lung volume. Furthermore, research on thermal imaging-based respiratory volume analysis remains limited. This study expands upon conventional respiratory rate-based approaches to provide a new direction for respiratory analysis using vision-based techniques.

Speech audiometry is a vital tool in assessing an individual’s ability to perceive and comprehend speech, traditionally requiring specialized testing that can be time-consuming and resource -intensive. This paper approaches a novel use of deep learning to predict speech audiometry using pure-tone audiometry (PTA) data. By utilizing PTA data, which measure hearing sensitivity at specific frequencies, we aim to develop a model that can bypass the need for direct speech testing. This study investigates two neural network architectures: a multi-layer perceptron (MLP) and a one-dimensional convolutional neural network (1D-CNN). These models are trained to predict key speech audiometry outcomes, including speech recognition thresholds and speech discrimination scores. To evaluate the effectiveness of these models, we employed two key performance metrics: the coefficient of determination (R²) and mean absolute error (MAE). The MLP model demonstrated predictive solid power with an R² score of 88.79% and an average MAE of 7.26, while the 1D-CNN model achieved a slightly higher level of accuracy with an MAE score of 88.35% and an MAE of 6.90. The superior performance of the 1D-CNN model suggests that it captures relevant features from PTA data more effectively than the MLP. These results show that both models hold promise for predicting speech audiometry, potentially simplifying the audiological evaluation process. This approach is applied in clinical settings for hearing loss assessment, the selection of hearing aids, and the development of personalized auditory rehabilitation programs.

Hearing loss significantly impacts daily communication, making accurate speech audiometry (SA) assessment essential for diagnosis and treatment. However, SA testing is time-consuming and resource-intensive, limiting its accessibility in clinical practice. This study aimed to develop a multi-class classification model that predicts SA results using pure-tone audiometry (PTA) data, enabling a more efficient and automated assessment. To achieve this, we implemented and compared MLP, RNN, gradient boosting, and XGBoost models, evaluating their performance using accuracy, F1 score, log loss, and confusion matrix analysis. Experimental results showed that gradient boosting achieved the highest accuracy, 86.22%, while XGBoost demonstrated a more balanced classification performance. The MLP achieved 85.77% and the RNN achieved 85.41%, exhibiting relatively low accuracy, with the RNN showing limitations due to the low temporal dependency of PTA data. Additionally, all models faced challenges predicting class 2 (borderline hearing levels) due to overlapping data distributions. These findings suggest that machine learning models, particularly gradient boosting and XGBoost, outperform deep learning models in SA prediction. Future research should focus on feature engineering, hyperparameter optimization, and ensemble approaches to enhance performance and validate real-world applicability. The proposed model could contribute to automating SA prediction and improving hearing assessment efficiency and patient care.