AI4COVID-19: AI Enabled Preliminary Diagnosis for COVID-19 from Cough Samples via an App (2004.01275v6)

Abstract: Background: The inability to test at scale has become humanity's Achille's heel in the ongoing war against the COVID-19 pandemic. A scalable screening tool would be a game changer. Building on the prior work on cough-based diagnosis of respiratory diseases, we propose, develop and test an AI-powered screening solution for COVID-19 infection that is deployable via a smartphone app. The app, named AI4COVID-19 records and sends three 3-second cough sounds to an AI engine running in the cloud, and returns a result within two minutes. Methods: Cough is a symptom of over thirty non-COVID-19 related medical conditions. This makes the diagnosis of a COVID-19 infection by cough alone an extremely challenging multidisciplinary problem. We address this problem by investigating the distinctness of pathomorphological alterations in the respiratory system induced by COVID-19 infection when compared to other respiratory infections. To overcome the COVID-19 cough training data shortage we exploit transfer learning. To reduce the misdiagnosis risk stemming from the complex dimensionality of the problem, we leverage a multi-pronged mediator centered risk-averse AI architecture. Results: Results show AI4COVID-19 can distinguish among COVID-19 coughs and several types of non-COVID-19 coughs. The accuracy is promising enough to encourage a large-scale collection of labeled cough data to gauge the generalization capability of AI4COVID-19. AI4COVID-19 is not a clinical grade testing tool. Instead, it offers a screening tool deployable anytime, anywhere, by anyone. It can also be a clinical decision assistance tool used to channel clinical-testing and treatment to those who need it the most, thereby saving more lives.

• The paper introduces an AI-based smartphone application that achieves over 90% accuracy in preliminary COVID-19 diagnosis by integrating deep transfer learning and multiple classifiers.
• It employs advanced audio processing techniques, converting cough sounds into Mel spectrograms and MFCC features to facilitate robust machine learning analysis.
• The system’s multi-tiered architecture supports remote screening and public health monitoring while significantly reducing false positive and negative rates.

AI4COVID-19: AI-Enabled Preliminary Diagnosis for COVID-19 from Cough Samples via an App

The paper presents an approach leveraging AI for the preliminary diagnosis of COVID-19 through cough analysis. Deployable as a smartphone application, this system seeks to fill the void of scalable, accessible testing methods that are particularly pertinent during pandemics with high transmission rates like COVID-19.

Methodology and Implementation

The AI4COVID-19 application records a user’s cough and processes the audio data using a cloud-based AI engine to deliver a diagnostic result within two minutes. The methodology harnesses the distinctive pathomorphological changes induced in the respiratory system by COVID-19, hypothesizing that these changes result in unique cough characteristics that are discernible to AI models.

Data Challenges and Pre-processing

A significant challenge was the dearth of labeled COVID-19 cough data. To mitigate this, the paper employs transfer learning, utilizing pre-trained models on similar tasks to improve the AI's diagnostic accuracy. Data pre-processing includes conversion to Mel spectrograms and extraction of features such as Mel Frequency Cepstral Coefficients (MFCC), transforming audio data into formats amenable to machine learning.

AI Architecture

The AI architecture is notably risk-averse and composed of three parallel classifiers:

1. Deep Transfer Learning-based Multi-Class (DTL-MC) Classifier - Uses a convolutional neural network (CNN) to distinguish COVID-19 from non-COVID-19 related coughs.
2. Classical Machine Learning-based Multi-Class (CML-MC) Classifier - Utilizes support vector machines drawing on manually extracted features.
3. Deep Transfer Learning-based Binary-Class (DTL-BC) Classifier - Focuses solely on binary classification between COVID-19 and non-COVID-19 coughs.

A mediator consolidates results from these classifiers, declaring a result as inconclusive if consensus is absent, thereby prioritizing the reduction of false diagnoses.

Results and Performance

The paper reports promising accuracy rates for the AI4COVID-19 system, with an emphasis on minimizing misdiagnosis. The approach achieves an overall classification accuracy exceeding 90% in some configurations, with the multi-pronged mediator approach noted for reducing both false positive and negative rates significantly.

Implications and Future Directions

This system offers several potential utilities:

• Remote Screening: Enables anytime, anywhere, individual-level preliminary screening, supporting overwhelmed healthcare systems and limiting pandemic spread.
• Complement to Clinical Tests: It is not intended to replace clinical-grade diagnostics but can serve as an adjunct tool, especially in resource-limited settings.
• Public Health Monitoring: By aggregating anonymized data, it can assist in broader epidemiological monitoring and analysis, aiding public health responses.

For future work, the paper suggests integrating additional data types such as breathing sounds and bio-markers to enhance diagnostic accuracy. Additionally, a greater diversity of cough samples from various other respiratory conditions is advised to strengthen model robustness.

Conclusion

The AI4COVID-19 provides an innovative, accessible alternative to traditional COVID-19 testing methods, capitalizing on AI's capability to discern subtle, condition-specific audio features. While the current model benefits from solid preliminary results, its practical efficacy will rely substantially on further development, extensive data collection, and real-world trials. The outlined framework exemplifies a pivotal step towards leveraging AI in global health crises, with implications extending beyond the immediate pandemic context.