Interpretable artificial intelligence framework for COVID-19 screening on chest X-rays

Abstract

COVID-19 has led to an unprecedented healthcare crisis with millions of infected people across the globe often pushing infrastructures, healthcare workers and entire economies beyond their limits. The scarcity of testing kits, even in developed countries, has led to extensive research efforts towards alternative solutions with high sensitivity. Chest radiological imaging paired with artificial intelligence (AI) can offer significant advantages in diagnosis of novel coronavirus infected patients. To this end, transfer learning techniques are used for overcoming the limitations emanating from the lack of relevant big datasets, enabling specialized models to converge on limited data, as in the case of X-rays of COVID-19 patients. In this study, we present an interpretable AI framework assessed by expert radiologists on the basis on how well the attention maps focus on the diagnostically-relevant image regions. The proposed transfer learning methodology achieves an overall area under the curve of 1 for a binary classification problem across a 5-fold training/testing dataset.

Keywords: COVID-19; chest X-rays; interpretable artificial intelligence; transfer learning.

Figure 1

Proposed architecture classifiers.

Figure 2

Average confusion matrix across all folds - Binary classification.

Figure 3

Average confusion matrix across all folds - Ternary classification.

Figure 4

Average confusion matrix across all folds - Quaternary classification.

Figure 5

Attention map of patient 23 - Binary classification - True positive COVID-19 with a certainty of 100% - Evaluated as grade 2 and 3 (left and right lungs, respectively) by the experts.

Figure 6

Attention map of patient 28 - Binary classification - True positive COVID-19 with a certainty of 100% - Evaluated as grade 3 (both left and right lungs) by the experts.

Figure 7

Attention map of patient 27 - Ternary classification - True positive COVID19 (predicted as COVID-19 with a certainty of 100%) - Evaluated as grade 0 and 4 (left and right lungs, respectively) by the experts.

Figure 8

Attention map of patient 8 - Ternary classification - False negative COVID-19 (predicted as pneumonia with a certainty of 73%) - Evaluated as grade 3 and 4 (left and right lungs, respectively) by the experts.

Figure 9

Attention map of patient 15 - Ternary classification - False Negative COVID-19 (predicted as pneumonia with a certainty of 53%) - Evaluated as grade 2 and 4 (left and right lungs, respectively) by the experts.

Figure 10

Attention map of patient 10 - Ternary classification - False negative COVID-19 (predicted as normal with a certainty of 95%) - Evaluated as grade 2 and 2 (left and right lungs, respectively) by the experts.

Figure 11

Attention map of patient 23 - Ternary classification - False negative COVID-19 (predicted as normal with a certainty of 98%) - Evaluated as grade 2 and 2 (left and right lungs, respectively) by the experts.