Automatic Cough Analysis for Non-Small Cell Lung Cancer Detection
Chiara Giangregorio, Cristina Maria Licciardello, Vanja Miskovic, Leonardo Provenzano, Alessandra Laura Giulia Pedrocchi, Andra Diana Dumitrascu, Arsela Prelaj, Marina Chiara Garassino, Emilia Ambrosini*, Simona Ferrante* · 2025
Read Automatic Cough Analysis for Non-Small Cell Lung Cancer Detection by Chiara Giangregorio, Cristina Maria Licciardello, Vanja Miskovic, Leonardo Provenzano, Alessandra Laura Giulia Pedrocchi, Andra Diana Dumitrascu, Arsela Prelaj, Marina Chiara Garassino, Emilia Ambrosini*, Simona Ferrante* (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-19174.txt.
Introduction
Chiara Giangregorio and co-authors test cough sound analysis for lung cancer screening in a 2025 study. Early detection of non-small cell lung cancer greatly improves patient outcomes. A cough recording is cheap and easy to collect, so it could serve as a first screen.
Recordings came from more than two hundred people, split between cancer patients and healthy controls. Machine learning tools such as support vector machines and XGBoost read the sounds. Deep learning with convolutional networks and a pretrained image model was also tried. The study weighs how well each method separates the two groups. A sound-based screen could flag who needs full testing first.
Catalog record
| CREATOR | Chiara Giangregorio, Cristina Maria Licciardello, Vanja Miskovic, Leonardo Provenzano, Alessandra Laura Giulia Pedrocchi, Andra Diana Dumitrascu, Arsela Prelaj, Marina Chiara Garassino, Emilia Ambrosini*, Simona Ferrante* |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 28 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-19174
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Automatic Cough Analysis for Non-Small Cell Lung Cancer Detection Chiara Giangregorio, Cristina Maria Licciardello, Vanja Miskovic, Leonardo Provenzano, Alessandra Laura Giulia Pedrocchi, Andra Diana Dumitrascu, Arsela Prelaj, Marina Chiara Garassino, Emilia Ambrosini*, Simona Ferrante* Abstract Early detection of non-small cell lung cancer (NSCLC) is critical for improving patient outcomes, and novel approaches are needed to facilitate early diagnosis. In this study, we explore the use of automatic cough analysis as a pre-screening tool for distinguishing between NSCLC patients and healthy controls. Cough audio recordings were prospectively acquired from a total of 227 subjects, divided into NSCLC patients and healthy controls. The recordings were analyzed using machine learning techniques, such as support vector machine (SVM) and XGBoost, as well as deep learning approaches, specifically convolutional neural networks (CNN) and transfer learning with VGG16. To enhance the interpretability of the machine learning model, we utilized Shapley Additive Explanations (SHAP). The fairness of the models across demographic groups was assessed by comparing the performance of the best model across different age groups ( ≤ \leq 58y and > > 58y) and gender using the equalized odds difference on the test set. The results demonstrate that CNN achieves the best performance, with an accuracy of 0.83 on the test set. Nevertheless, SVM achieves slightly lower performances (accuracy of 0.76 in validation and 0.78 in the test set), making it suitable in contexts with low computational power. The use of SHAP for SVM interpretation further enhances model transparency, making it more trustworthy for clinical applications. Fairness analysis shows slightly higher disparity across age (0.15) than gender (0.09) on the test set. Therefore, to strengthen our findings’ reliability, a larger, more diverse, and unbiased dataset is needed—particularly including individuals at risk of NSCLC and those in early disease stages. Clinical relevance — This study highlights the promise of integrating automatic cough analysis with machine learning techniques for improving lung cancer screening methods. I Introduction Worldwide, lung cancer is one of the most frequently diagnosed cancers and the leading cause of cancer-related death [ 1 ] . It is classified into two main types: non-small cell lung cancer (NSCLC) and small cell lung cancer (SCLC), based on their histopathological characteristics and cellular morphology under microscopic examination [ 2 ] . About 85% of the cases are classified as NSCLC, which usually progresses more slowly than SCLC [ 3 ] . However, by the time NSCLC is diagnosed, 40% of cases have already spread beyond the lungs [ 4 ] . Detecting NSCLC at an early stage provides the most promising prognosis. However, diagnosing NSCLC remains challenging, as its symptoms often mimic those of more common diseases or the long-term effects of smoking [ 5 ] . Low-dose chest CT (LDCT) has been shown to reduce mortality by enabling early detection; however, its widespread implementation requires screening a massive population, exposing individuals to radiation from X-rays. Additionally, traditional imaging methods can be costly and often remain inaccessible for routine screening. Consequently, 80% of NSCLC cases are diagnosed at advanced stages, significantly complicating treatment efforts and reducing survival rates [ 6 ] . This highlights the need for alternative approaches that are both scalable and affordable for early and accessible detection. Cough, a common but often neglected symptom of lung cancer, offers an opportunity as a digital biomarker. While artificial intelligence (AI) has been successfully applied to classify cough in conditions such as COVID-19, pertussis, bronchitis, or pneumonia [ 7 ] , its potential for detecting NSCLC remains largely underexplored. Indeed, the potential integration of AI-driven cough analysis into everyday devices like smartphones would offer a real-time, cost-effective, and scalable solution for respiratory disease screening [ 8 , 9 , 10 ] . During the COVID-19 pandemic, AI-based cough classification gained significant attention, with researchers leveraging machine learning (ML) models to enhance the speed and efficiency of screening efforts [ 11 , 12 , 13 , 14 , 15 , 16 ] . One significant study presented by Pahar et al. [ 17 ] proposed a machine learning-based COVID-19 cough classifier that could discriminate between COVID-19 positive coughs, COVID-19 negative coughs, and healthy coughs, all recorded on smartphones, achieving an area under the ROC curve (AUC) of 0.98. Moreover, research carried out by Bales et al. [ 18 ] demonstrated how machine learning models and convolutional neural networks (CNNs) can effectively detect various respiratory conditions, such as bronchitis and pertussis, based on cough sounds. The study achieved over 89% accuracy in both cough detection and disease classification, showcasing the versatility and potential of ML in diagnosing a range of respiratory conditions. Similarly, Google researchers presented their Health Acoustic Representations (HeAR) system, which leverages large-scale self-supervised learning (SSL) to process health-related acoustic signals, such as coughs and breaths. Trained on a vast dataset of over 313 million two-second audio clips, HeAR aims to create a general-purpose audio encoder that excels across a range of health acoustic tasks, including cough classification and lung function analysis, such as spirometry. This research highlights the potential of SSL to enhance generalization in health acoustic tasks, offering potential scalable and accessible solutions for healthcare [ 19 ] . In this context, the primary objective of this study is to investigate the feasibility of employing cough as a digital biomarker to differentiate between healthy individuals and NSCLC patients. Specifically, the study explores the potential of ML and deep learning (DL) techniques to classify forced cough sounds, offering a non-invasive approach to lung cancer detection. By applying AI techniques, this work seeks to contribute to understanding how computational methods can support personalized medicine, trying to identify potential improvements in the detection and management of lung cancer through non-invasive methods. II Methods II-A Data Collection and Pre-processing Forced cough recordings were collected from two groups: healthy control subjects (Healthy group) and patients with stage IIIB-IV NSCLC undergoing immunotherapy (Cancer group). For each participant, a single cough recording, along with their age and sex and smoking habits, was recorded. Participants signed written informed consent prior to inclusion in the study. The study was conducted as part of the EU-horizon project I3LUNG (NCT05537922) and was approved by the Ethical Committee of Fondazione IRCCS Istituto Nazionale dei Tumori di Milano (approval number 147/22, approved on 04/11/2024). Audio recordings were acquired using a mobile app or a web app, with a sampling frequency of 16 kHz, and were stored on an Amazon Web Server. Pre-processing steps were implemented to address potential inconsistencies in audio quality due to varying recording devices and environmental conditions, including loudness normalization, low-pass filtering, and downsampling to 12kHz. After preprocessing, segmentation was performed to isolate meaningful cough events based on the cough segmentation algorithm developed by Orlandic et al. [ 20 ] . The recordings were then divided into 90%, 10% for training, and testing, respectively. The test set was used exclusively to evaluate the performance of the final models. Mann-Whitney test for independent samples was applied to compare the two groups in terms of age. A Pearson Chi-squared test was instead used for biological sex and smoking habits. Subsequently, two parallel feature extraction and analysis pipelines were carried out as inputs for classical machine learning and deep and transfer learning models. II-B Machine Learning approach II-B 1