Interpretability Analysis of Deep Models for COVID-19 Detection

TL;DR

This paper analyzes how deep convolutional neural networks detect COVID-19 from audio data, focusing on interpretability, feature importance, and model decision processes, demonstrating high accuracy and robustness to spurious data.

Contribution

It introduces an interpretability framework for CNN-based COVID-19 detection in audio, highlighting feature importance and model attention, with robust performance despite data biases.

Findings

Models achieve 94.44% accuracy in COVID-19 detection.

Spectrograms and prosodic features are key for model decisions.

Heat maps reveal focus on high-energy spectrogram regions.

Abstract

During the outbreak of COVID-19 pandemic, several research areas joined efforts to mitigate the damages caused by SARS-CoV-2. In this paper we present an interpretability analysis of a convolutional neural network based model for COVID-19 detection in audios. We investigate which features are important for model decision process, investigating spectrograms, F0, F0 standard deviation, sex and age. Following, we analyse model decisions by generating heat maps for the trained models to capture their attention during the decision process. Focusing on a explainable Inteligence Artificial approach, we show that studied models can taken unbiased decisions even in the presence of spurious data in the training set, given the adequate preprocessing steps. Our best model has 94.44% of accuracy in detection, with results indicating that models favors spectrograms for the decision process, particularly, high energy areas in the spectrogram related to prosodic domains, while F0 also leads to efficient COVID-19 detection.