Search

A recent spate of state-of-the-art semi- and un-supervised solutions disentangle and encode image "content" into a spatial tensor and image appearance or "style" into a vector, to achieve good performance in spatially equivariant tasks (e.g. image-to-image translation). To achieve this, they employ different model design, learning objective, and data biases. While considerable effort has been made to measure disentanglement in vector representations, and assess its impact on task performance, such analysis for (spatial) content-style disentanglement is lacking. In this paper, we conduct an empirical study to investigate the role of different biases in content-style disentanglement settings and unveil the relationship between the degree of disentanglement and task performance. In particular, we consider the setting where we: (i) identify key design choices and learning constraints for three popular content-style disentanglement models; (ii) relax or remove such constraints in an ablation fashion; and (iii) use two metrics to measure the degree of disentanglement and assess its effect on each task performance. Our experiments reveal that there is a “sweet spot” between disentanglement, task performance and - surprisingly – content interpretability, suggesting that blindly forcing for higher disentanglement can hurt model performance and content factors semanticness. Our findings, as well as the used task-independent metrics, can be used to guide the design and selection of new models for tasks where content-style representations are useful.

Thanks to their ability to learn data distributions without requiring paired data, Generative Adversarial Networks (GANs) have become an integral part of many computer vision methods, including those developed for medical image segmentation. These methods jointly train a segmentor and an adversarial mask discriminator, which provides a data-driven shape prior. At inference, the discriminator is discarded, and only the segmentor is used to predict label maps on test images. But should we discard the discriminator? Here, we argue that the life cycle of adversarial discriminators should not end after training. On the contrary, training stable GANs produces powerful shape priors that we can use to correct segmentor mistakes at inference. To achieve this, we develop stable mask discriminators that do not overfit or catastrophically forget. At test time, we fine-tune the segmentor on each individual test instance until it satisfies the learned shape prior. Our method is simple to implement and increases model performance. Moreover, it opens new directions for re-using mask discriminators at inference. We release the code used for the experiments at https://vios-s.github.io/adversarial-test-time-training., Comment: Accepted at: Domain Adaptation and Representation Transfer (DART) 2021

Thanks to their ability to learn flexible data-driven losses, Generative Adversarial Networks (GANs) are an integral part of many semi- and weakly-supervised methods for medical image segmentation. GANs jointly optimise a generator and an adversarial discriminator on a set of training data. After training is complete, the discriminator is usually discarded, and only the generator is used for inference. But should we discard discriminators? In this work, we argue that training stable discriminators produces expressive loss functions that we can re-use at inference to detect and \textit{correct} segmentation mistakes. First, we identify key challenges and suggest possible solutions to make discriminators re-usable at inference. Then, we show that we can combine discriminators with image reconstruction costs (via decoders) to endow a causal perspective to test-time training and further improve the model. Our method is simple and improves the test-time performance of pre-trained GANs. Moreover, we show that it is compatible with standard post-processing techniques and it has the potential to be used for Online Continual Learning. With our work, we open new research avenues for re-using adversarial discriminators at inference. Our code is available at https://vios-s.github.io/adversarial-test-time-training.

A recent spate of state-of-the-art semi- and un-supervised solutions disentangle and encode image "content" into a spatial tensor and image appearance or "style" into a vector, to achieve good performance in spatially equivariant tasks (e.g. image-to-image translation). To achieve this, they employ different model design, learning objective, and data biases. While considerable effort has been made to measure disentanglement in vector representations, and assess its impact on task performance, such analysis for (spatial) content - style disentanglement is lacking. In this paper, we conduct an empirical study to investigate the role of different biases in content-style disentanglement settings and unveil the relationship between the degree of disentanglement and task performance. In particular, we consider the setting where we: (i) identify key design choices and learning constraints for three popular content-style disentanglement models; (ii) relax or remove such constraints in an ablation fashion; and (iii) use two metrics to measure the degree of disentanglement and assess its effect on each task performance. Our experiments reveal that there is a "sweet spot" between disentanglement, task performance and - surprisingly - content interpretability, suggesting that blindly forcing for higher disentanglement can hurt model performance and content factors semanticness. Our findings, as well as the used task-independent metrics, can be used to guide the design and selection of new models for tasks where content-style representations are useful., Comment: 28 pages, 10 figures

There has been an increasing focus in learning interpretable feature representations, particularly in applications such as medical image analysis that require explainability, whilst relying less on annotated data (since annotations can be tedious and costly). Here we build on recent innovations in style-content representations to learn anatomy, imaging characteristics (appearance) and temporal correlations. By introducing a self-supervised objective of predicting future cardiac phases we improve disentanglement. We propose a temporal transformer architecture that given an image conditioned on phase difference, it predicts a future frame. This forces the anatomical decomposition to be consistent with the temporal cardiac contraction in cine MRI and to have semantic meaning with less need for annotations. We demonstrate that using this regularization, we achieve competitive results and improve semi-supervised segmentation, especially when very few labelled data are available. Specifically, we show Dice increase of up to 19\% and 7\% compared to supervised and semi-supervised approaches respectively on the ACDC dataset. Code is available at: https://github.com/gvalvano/sdtnet .

There has been an increasing focus in learning interpretable feature representations, particularly in applications such as medical image analysis that require explainability, whilst relying less on annotated data (since annotations can be tedious and costly). Here we build on recent innovations in style-content representations to learn anatomy, imaging characteristics (appearance) and temporal correlations. By introducing a self-supervised objective of predicting future cardiac phases we improve disentanglement. We propose a temporal transformer architecture that given an image conditioned on phase difference, it predicts a future frame. This forces the anatomical decomposition to be consistent with the temporal cardiac contraction in cine MRI and to have semantic meaning with less need for annotations. We demonstrate that using this regularization, we achieve competitive results and improve semi-supervised segmentation, especially when very few labelled data are available. Specifically, we show Dice increase of up to 19\% and 7\% compared to supervised and semi-supervised approaches respectively on the ACDC dataset. Code is available at: https://github.com/gvalvano/sdtnet ., Comment: 9 pages, 4 figures (1 .gif), 1 table

In Europe the 20% of the CT scans cover the thoracic region. The acquired images contain information about the cardiovascular system that often remains latent due to the lack of contrast in the cardiac area. On the other hand, the contrast enhanced computed tomography (CECT) represents an imaging technique that allows to easily assess the cardiac chambers volumes and the contrast dynamics. With this work we aim to face the problem of extraction and presentation of these latent information, using a deep learning approach with convolutional neural networks. Starting from the extraction of relevant features from the image without contrast medium, we try to re-map them on features typical of CECT, to synthesize an image characterized by an attenuation in the cardiac chambers as if a virtually iodine contrast medium was injected. The purposes are to guarantee an estimation of the left cardiac chambers volume and to perform an evaluation of the contrast dynamics. Our approach is based on a deconvolutional network trained on a set of 120 patients who underwent both CT acquisitions in the same contrastographic arterial phase and the same cardiac phase. To ensure a reliable predicted CECT image, in terms of values and morphology, a custom loss function is defined by combining an error function to find a pixel-wise correspondence, which takes into account the similarity in term of Hounsfield units between the input and output images and by a cross-entropy computed on the binarized versions of the synthesized and of the real CECT image. The proposed method is finally tested on 20 subjects., 8 pages,3 figures

Skull-stripping methods aim to remove the non-brain tissue from acquisition of brain scans in magnetic resonance (MR) imaging. Although several methods sharing this common purpose have been presented in literature, they all suffer from the great variability of the MR images. In this work we propose a novel approach based on Convolutional Neural Networks to automatically perform the brain extraction obtaining cutting-edge performance in the NFBS public database. Additionally, we focus on the efficient training of the neural network designing an effective data augmentation pipeline. Obtained results are evaluated through Dice metric, obtaining a value of 96.5%, and processing time, with 4.5s per volume., Comment: April 2018. Content: 8 pages, 2 figures

Recent research put a big effort in the development of deep learning architectures and optimizers obtaining impressive results in areas ranging from vision to language processing. However little attention has been addressed to the need of a methodological process of data collection. In this work we hypothesize that high quality data for supervised learning can be selected in an unsupervised manner and that by doing so one can obtain models capable to generalize better than in the case of random training set construction. However, preliminary results are not robust and further studies on the subject should be carried out., Comment: Technical Report --- 8 pages, 3 figures New tests demonstrated that the system, as is, is not able to create reproducible results. Further study on the topic should be done