Bias Analysis on Public X-Ray Image Datasets of Pneumonia and COVID-19 Patients

Abstract

Chest X-ray images are useful for early COVID-19 diagnosis with the advantage that X-ray devices are already available in health centers and images are obtained immediately. Some datasets containing X-ray images with cases (pneumonia or COVID-19) and controls have been made available to develop machine-learning-based methods to aid in diagnosing the disease. However, these datasets are mainly composed of different sources coming from pre-COVID-19 datasets and COVID-19 datasets. Particularly, we have detected a significant bias in some of the released datasets used to train and test diagnostic systems, which might imply that the results published are optimistic and may overestimate the actual predictive capacity of the techniques proposed. In this paper, we analyze the existing bias in some commonly used datasets and propose a series of preliminary steps to carry out before the classic machine learning pipeline in order to detect possible biases, to avoid them if possible and to report results that are more representative of the actual predictive power of the methods under analysis.

Explainable Artificial Intelligence for Bias Detection in COVID CT-Scan Classifiers

“…As a counterpoint to DenseNet models, it would be interesting to study other representative state-of-the-art architectures and compare the XAI heatmaps generated by their classifications. The VGG architecture [ 17 ], namely, VGG16, was chosen for this purpose, because it has been previously probed for CXR classification biases [ 8 ] and it is a commonly used high performance classifier for chest radiography type images in general [ 18 , 19 , 20 ].…”

“…Although it seems to be generally agreed upon in literature that CT-Scans allow for better discernment and diagnosis for this particular disease [ 1 , 8 ], the majority of works focuses on CXR, perhaps due to the more immediate availability of CXR in general.…”

“…As of yet, few publications seem to address this problem beyond passing comments. As a counterpoint, Catala et al [ 8 ] study the specific case of a VGG16 classifier [ 9 ] trained on a CXR datasets. The author used GradCAM [ 10 ] heatmaps to visualize biases in classifications.…”

3D Reconstruction of Laparoscope Images With Contrastive Learning Methods

“…Method [17] scored similar to ours, but it was easy to fall into local optimum. Method [18] also tends to fall into local optimum with low score. Our method scores higher in true depth and is more selective.…”

“…The top line is the original frame image input. [17][18][3], respectively, the current 3d reconstruction method of endoscopic image; the fifth act is the reconstruction result of our method; [19][20][21] are the reconstruction result of the computational network in this model replaced by other deep learning methods Since the model proposed by [17], [18] and [3] was designed and trained for endoscopic images, the reconstruction results were better than those of [19], [20] and [21], but some key feature points were still missing. Our reconstruction results could better extract most of the feature points, and the reconstruction effect was good In our method, endoscope has the best effect, because the model is trained according to the data of endoscope, and our method has the worst effect of arthroscopy reconstruction.…”