Interpretability Analysis of Deep Models for COVID-19 Detection
Authors:
Daniel Peixoto Pinto da Silva,
Edresson Casanova,
Lucas Rafael Stefanel Gris,
Arnaldo Candido Junior,
Marcelo Finger,
Flaviane Svartman,
Beatriz Raposo,
Marcus Vinícius Moreira Martins,
Sandra Maria Aluísio,
Larissa Cristina Berti,
João Paulo Teixeira
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.…
▽ More
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.
△ Less
Submitted 25 November, 2022;
originally announced November 2022.
