No abstract

One of the most rapidly advancing areas of deep learning research aims at creating models that learn to disentangle the latent factors of variation from a data distribution. However, modeling joint probability mass functions is usually prohibitive, which motivates the use of conditional models assuming that some information is given as input. In the domain of numerical cognition, deep learning architectures have successfully demonstrated that approximate numerosity representations can emerge in multi-layer networks that build latent representations of a set of images with a varying number of items. However, existing models have focused on tasks requiring to conditionally estimate numerosity information from a given image. Here, we focus on a set of much more challenging tasks, which require to conditionally generate synthetic images containing a given number of items. We show that attention-based architectures operating at the pixel level can learn to produce well-formed images approximately containing a specific number of items, even when the target numerosity was not present in the training distribution.

No abstract

The lack of large labeled medical imaging datasets, along with significant inter-individual variability compared to clinically established disease classes, poses significant challenges in exploiting medical imaging information in a precision medicine paradigm, where in principle dense patient-specific data can be employed to formulate individual predictions and/or stratify patients into finer-grained groups which may follow more homogeneous trajectories and therefore empower clinical trials. In order to efficiently explore the effective degrees of freedom underlying variability in medical images in an unsupervised manner, in this work we propose an unsupervised autoencoder framework which is augmented with a contrastive loss to encourage high separability in the latent space. The model is validated on (medical) benchmark datasets. As cluster labels are assigned to each example according to cluster assignments, we compare performance with a supervised transfer learning baseline. Our methods achieves similar performance to the supervised architecture, indicating that separation in the latent space reproduces expert medical observer-assigned labels. The proposed method could be beneficial for patient stratification, exploring new subdivision of larger classes or pathological continua or, due to its sampling abilities in a variation setting, data augmentation in medical image processing.

Positron emission tomography (PET) is a noninvasive imaging technology able to assess the metabolic or functional state of healthy and/or pathological tissues. In clinical practice, PET data are usually acquired statically and normalized for the evaluation of the standardized uptake value (SUV). In contrast, dynamic PET acquisitions provide information about radiotracer delivery to tissue, its interaction with the target, and its physiological washout. The shape of the time activity curves (TACs) embeds tissue-specific biochemical properties. Conventionally, TACs are employed along with information about blood plasma activity concentration, i.e., the arterial input function, and tracer-specific compartmental models to obtain a full quantitative analysis of PET data. This method's primary disadvantage is the requirement for invasive arterial blood sample collection throughout the whole PET scan. In this study, we employ a variety of deep learning models to illustrate the diagnostic potential of dynamic PET acquisitions of varying lengths for discriminating breast cancer lesions in the absence of arterial blood sampling compared to static PET only. Our findings demonstrate that the use of TACs, even in the absence of arterial blood sampling and even when using only a share of all timeframes available, outperforms the discriminative ability of conventional SUV analysis.