Abstract
Chest radiography (CXR) is one of the most used medical imaging modalities. Nevertheless, the interpretation of CXR images is time-consuming and subject to variability. As such, automated systems for pathology detection have been proposed and promising results have been obtained, particularly using deep learning. However, these tools suffer from poor explainability, which represents a major hurdle for their adoption in clinical practice. One proposed explainability method in CXR is through contrastive examples, i.e. by showing an alternative version of the CXR except without the lesion being investigated. While image-level normal/healthy image synthesis has been explored in literature, normal patch synthesis via inpainting has received little attention. In this work, a method to synthesize contrastive examples in CXR based on local synthesis of normal CXR patches is proposed. Based on a contextual attention inpainting network (CAttNet), an anatomically-guided inpainting network (AnaCAttNet) is proposed that leverages anatomical information of the original CXR through segmentation to guide the inpainting for a more realistic reconstruction. A quantitative evaluation of the inpainting is performed, showing that AnaCAttNet outperforms CAttNet (FID of 0.0125 and 0.0132 respectively). Qualitative evaluation by three readers also showed that AnaCAttNet delivers superior reconstruction quality and anatomical realism. In conclusion, the proposed anatomical segmentation module for inpainting is shown to improve inpainting performance.

Abstract
The increase in complexity of deep learning models demands explanations that can be obtained with methods like Grad-CAM. This method computes an importance map for the last convolutional layer relative to a specific class, which is then upsampled to match the size of the input. However, this final step assumes that there is a spatial correspondence between the last feature map and the input, which may not be the case. We hypothesize that, for models with large receptive fields, the feature spatial organization is not kept during the forward pass, which may render the explanations devoid of meaning. To test this hypothesis, common architectures were applied to a medical scenario on the public VinDr-CXR dataset, to a subset of ImageNet and to datasets derived from MNIST. The results show a significant dispersion of the spatial information, which goes against the assumption of Grad-CAM, and that explainability maps are affected by this dispersion. Furthermore, we discuss several other caveats regarding Grad-CAM, such as feature map rectification, empty maps and the impact of global average pooling or flatten layers. Altogether, this work addresses some key limitations of Grad-CAM which may go unnoticed for common users, taking one step further in the pursuit for more reliable explainability methods.

Abstract
Chest X-ray scans are frequently requested to detect the presence of abnormalities, due to their low-cost and non-invasive nature. The interpretation of these images can be automated to prioritize more urgent exams through deep learning models, but the presence of image artifacts, e.g. lettering, often generates a harmful bias in the classifiers and an increase of false positive results. Consequently, healthcare would benefit from a system that selects the thoracic region of interest prior to deciding whether an image is possibly pathologic. The current work tackles this binary classification exercise, in which an image is either normal or abnormal, using an attention-driven and spatially unsupervised Spatial Transformer Network (STERN), that takes advantage of a novel domain-specific loss to better frame the region of interest. Unlike the state of the art, in which this type of networks is usually employed for image alignment, this work proposes a spatial transformer module that is used specifically for attention, as an alternative to the standard object detection models that typically precede the classifier to crop out the region of interest. In sum, the proposed end-to-end architecture dynamically scales and aligns the input images to maximize the classifier's performance, by selecting the thorax with translation and non-isotropic scaling transformations, and thus eliminating artifacts. Additionally, this paper provides an extensive and objective analysis of the selected regions of interest, by proposing a set of mathematical evaluation metrics. The results indicate that the STERN achieves similar results to using YOLO-cropped images, with reduced computational cost and without the need for localization labels. More specifically, the system is able to distinguish abnormal frontal images from the CheXpert dataset, with a mean AUC of 85.67% -a 2.55% improvement vs. the 0.98% improvement achieved by the YOLO-based counterpart in comparison to a standard baseline classifier. At the same time, the STERN approach requires less than 2/3 of the training parameters, while increasing the inference time per batch in less than 2 ms. Code available via GitHub.

Abstract
Machine Learning models need large amounts of annotated data for training. In the field of medical imaging, labeled data is especially difficult to obtain because the annotations have to be performed by qualified physicians. Natural Language Processing (NLP) tools can be applied to radiology reports to extract labels for medical images automatically. Compared to manual labeling, this approach requires smaller annotation efforts and can therefore facilitate the creation of labeled medical image data sets. In this article, we summarize the literature on this topic spanning from 2013 to 2023, starting with a meta-analysis of the included articles, followed by a qualitative and quantitative systematization of the results. Overall, we found four types of studies on the extraction of labels from radiology reports: those describing systems based on symbolic NLP, statistical NLP, neural NLP, and those describing systems combining or comparing two or more of the latter. Despite the large variety of existing approaches, there is still room for further improvement. This work can contribute to the development of new techniques or the improvement of existing ones.

Abstract
Artificial intelligence (AI) models have received considerable attention in recent years for their ability to identify optical coherence tomography (OCT) biomarkers with clinical diagnostic potential and predict disease progression. This study aims to externally validate a deep learning (DL) algorithm by comparing its segmentation of retinal layers and fluid with a gold-standard method for manually adjusting the automatic segmentation of the Heidelberg Spectralis HRA + OCT software Version 6.16.8.0. A total of sixty OCT images of healthy subjects and patients with intermediate and exudative age-related macular degeneration (AMD) were included. A quantitative analysis of the retinal thickness and fluid area was performed, and the discrepancy between these methods was investigated. The results showed a moderate-to-strong correlation between the metrics extracted by both software types, in all the groups, and an overall near-perfect area overlap was observed, except for in the inner segment ellipsoid (ISE) layer. The DL system detected a significant difference in the outer retinal thickness across disease stages and accurately identified fluid in exudative cases. In more diseased eyes, there was significantly more disagreement between these methods. This DL system appears to be a reliable method for accessing important OCT biomarkers in AMD. However, further accuracy testing should be conducted to confirm its validity in real-world settings to ultimately aid ophthalmologists in OCT imaging management and guide timely treatment approaches.