Within a legal framework, fairness in datasets and models is typically assessed by dividing observations into predefined groups and then computing fairness measures (e.g., Disparate Impact or Equality of Odds with respect to gender). However, when sensitive attributes such as skin color are continuous, dividing into default groups may overlook or obscure the discrimination experienced by certain minority subpopulations. To address this limitation, we propose a fairness-based grouping approach for continuous (possibly multidimensional) sensitive attributes. By grouping data according to observed levels of discrimination, our method identifies the partition that maximizes a novel criterion based on inter-group variance in discrimination, thereby isolating the most critical subgroups. We validate the proposed approach using multiple synthetic datasets and demonstrate its robustness under changing population distributions—revealing how discrimination is manifested within the space of sensitive attributes. Furthermore, we examine a specialized setting of monotonic fairness for the case of skin color. Our empirical results on both CelebA and FFHQ, leveraging the skin tone as predicted by an industrial proprietary algorithm, show that the proposed segmentation uncovers more nuanced patterns of discrimination than previously reported, and that these findings remain stable across datasets for a given model. Finally, we leverage our grouping model for debiasing purpose, aiming at predicting fair scores with group-by-group post-processing. The results demonstrate that our approach improves fairness while having minimal impact on accuracy, thus confirming our partition method and opening the door for industrial deployment.

Counterfactual frameworks have grown popular in machine learning for both explaining algorithmic decisions but also defining individual notions of fairness, more intuitive than typical group fairness conditions. However, state-of-the-art models to compute counterfactuals are either unrealistic or unfeasible. In particular, while Pearl's causal inference provides appealing rules to calculate counterfactuals, it relies on a model that is unknown and hard to discover in practice. We address the problem of designing realistic and feasible counterfactuals in the absence of a causal model. We define transport-based counterfactual models as collections of joint probability distributions between observable distributions, and show their connection to causal counterfactuals. More specifically, we argue that optimal-transport theory defines relevant transport-based counterfactual models, as they are numerically feasible, statistically-faithful, and can coincide under some assumptions with causal counterfactual models. Finally, these models make counterfactual approaches to fairness feasible, and we illustrate their practicality and efficiency on fair learning. With this paper, we aim at laying out the theoretical foundations for a new, implementable approach to counterfactual thinking.

In this article, we explore the use of novel neural-network architectures to distinguish natural seepages from artificial slicks in synthetic aperture radar (SAR) images. They exploit a distinctive property of natural seepages, which is their temporal recurrence in the same geographical area. This information can be captured in different SAR images acquired at the same location over time, but not necessarily at a regular time frequency. The proposed neural-network architectures are then built as specific block layers, which efficiently treat the unordered temporal information, followed by more conventional neural-network layers, which are widely used for image classification. Different block layers for unordered temporal information are compared on Sentinel-1 images acquired in the Aegean Sea. Following data augmentation steps, our dataset contains a consistent subset of images gathered among 16000 time-patches. We demonstrate that nonlinear time-based block layers and block layers that avoid information bottlenecks are most efficient to discriminate between natural and artificial oil spills. Compared with standard neural-networks which use single time frames, the integration of unordered temporal information increases the overall accuracy (OA) from 82% to 92% on our dataset, which demonstrates the effectiveness of the proposed approach.

Predicting tumor growth and its response to therapy remains a major challenge in cancer research and strongly relies on tumor growth models. In this paper, we introduce, calibrate, and verify a novel image-driven reaction-diffusion model of avascular tumor growth. The model allows for proliferation, death and spread of tumor cells, and accounts for nutrient distribution and hypoxia. It is constrained by longitudinal time series of dynamic contrast-enhancement-MRI images. Tumor specific parameters are estimated from two early time points and used to predict the spatio-temporal evolution of the tumor volume and cell densities at later time points. We first test our parameter estimation approach on synthetic data from 15 generated tumors. Our in silico study resulted in small volume errors (<5%) and high Dice overlaps (>97%), showing that model parameters can be successfully recovered and used to accurately predict the tumor growth. Encouraged by these results, we apply our model to seven pre-clinical cases of breast carcinoma. We are able to show promising preliminary results, especially for the estimation for early time points. Processes like angiogenesis and apoptosis should be included to further improve predictions for later time points.

An algorithm dedicated to automatic segmentation of breast magnetic resonance images is presented in this paper. Our approach is based on a pipeline that includes a denoising step and statistical segmentation. The noise removal preprocessing relies on an anisotropic diffusion scheme, whereas the statistical segmentation is conducted through a Markov random field model. The continuous updating of all parameters governing the diffusion process enables automatic denoising, and the partial volume effect is also addressed during the labeling step. To assess the relevance, the Jaccard similarity coefficient was computed. Experiments were conducted on synthetic data and breast magnetic resonance images extracted from a high-risk population. The relevance of the approach for the dataset is highlighted, and we demonstrate accuracy superior to that of traditional clustering algorithms. The results emphasize the benefits of both denoising guided by input data and the inclusion of spatial dependency through a Markov random field. For example, the Jaccard coefficient for the clinical data was increased by 114%, 109%, and 140% with respect to a K-means algorithm and, respectively, for the adipose, glandular and muscle and skin components. Moreover, the agreement between the manual segmentations provided by an experienced radiologist and the automatic segmentations performed with this algorithm was good, with Jaccard coefficients equal to 0.769, 0.756, and 0.694 for the above-mentioned classes.

Real-time ultrasound image acquisition is a pivotal resource in the medical community, in spite of its limited image quality. This poses challenges to image registration methods, particularly to those driven by intensity values. We address these difficulties in a novel diffeomorphic registration technique for tumor tracking in series of 2-D liver ultrasound. Our method has two main characteristics: 1) each voxel is described by three image features: intensity, local phase, and phase congruency; 2) we compute a set of forces from either local information (Demons-type of forces), or spatial correspondences supplied by a block-matching scheme, from each image feature. A family of update deformation fields which are defined by these forces, and inform upon the local or regional contribution of each image feature are then composed to form the final transformation. The method is diffeomorphic, which ensures the invertibility of deformations. The qualitative and quantitative results yielded by both synthetic and real clinical data show the suitability of our method for the application at hand.

In the framework of large deformation diffeomorphic metric mapping (LDDMM), we present a practical methodology to integrate prior knowledge about the registered shapes in the regularizing metric. Our goal is to perform rich anatomical shape comparisons from volumetric images with the mathematical properties offered by the LDDMM framework. We first present the notion of characteristic scale at which image features are deformed. We then propose a methodology to compare anatomical shape variations in a multi-scale fashion, i.e., at several characteristic scales simultaneously. In this context, we propose a strategy to quantitatively measure the feature differences observed at each characteristic scale separately. After describing our methodology, we illustrate the performance of the method on phantom data. We then compare the ability of our method to segregate a group of subjects having Alzheimer's disease and a group of controls with a classical coarse to fine approach, on standard 3D MR longitudinal brain images. We finally apply the approach to quantify the anatomical development of the human brain from 3D MR longitudinal images of pre-term babies. Results show that our method registers accurately volumetric images containing feature differences at several scales simultaneously with smooth deformations.

Within-subject analysis in fMRI essentially addresses two problems, the detection of brain regions eliciting evoked activity and the estimation of the underlying dynamics. In Makni et aL, 2005 and Makni et aL, 2008, a detection-estimation framework has been proposed to tackle these problems jointly, since they are connected to one another. In the Bayesian formalism, detection is achieved by modeling activating and nonactivating voxels through independent mixture models (IMM) within each region while hemodynamic response estimation is performed at a regional scale in a nonparametric way. Instead of IMMs, in this paper we take advantage of spatial mixture models (SMM) for their nonlinear spatial regularizing properties. The proposed method is unsupervised and spatially adaptive in the sense that the amount of spatial correlation is automatically tuned from the data and this setting automatically varies across brain regions. In addition, the level of regularization is specific to each experimental condition since both the signal-to-noise ratio and the activation pattern may vary across stimulus types in a given brain region. These aspects require the precise estimation of multiple partition functions of underlying Ising fields. This is addressed efficiently using first path sampling for a small subset of fields and then using a recently developed fast extrapolation technique for the large remaining set. Simulation results emphasize that detection relying on supervised SMM outperforms its IMM counterpart and that unsupervised spatial mixture models achieve similar results without any hand-tuning of the correlation parameter. On real datasets, the gain is illustrated in a localizer fMRI experiment: brain activations appear more spatially resolved using SMM in comparison with classical general linear model (GLM)-based approaches, while estimating a specific parcel-based HRF shape. Our approach therefore validates the treatment of unsmoothed fMRI data without fixed GLM definition at the subject level and makes also the classical strategy of spatial Gaussian filtering deprecated.

In this paper, a novel statistical parcellation of intra-subject functional MRI (fMRI) data is proposed. The key idea is to identify functionally homogenous regions of interest from their hemodynamic parameters. To this end, a non-parametric voxel-based estimation of hemodynamic response function is performed as a prerequisite. Then, the extracted hemodynamic features are entered as the input data of a Multivariate Spatial Gaussian Mixture Model (MSGMM) to be fitted. The goal of the spatial aspect is to favor the recovery of connected components in the mixture. Our statistical clustering approach is original in the sense that it extends existing works done on univariate spatially regularized Gaussian mixtures. A specific Gibbs sampler is derived to account for different covariance structures in the feature space. On realistic artificial fMRI datasets, it is shown that our algorithm is helpful for identifying a parsimonious functional parcellation required in the context of joint detection-estimation of brain activity. This allows us to overcome the classical assumption of spatial stationarity of the BOLD signal model.

Symmetric ising models define the simplest discrete Markov random fields that can be used for segmentation purpose. Unsupervised segmentation requires an automatic setting of the temperature parameter of ising fields. To this end, partition function (PF) estimation becomes a key issue. In this paper, we present a bilinear extrapolation technique for a fast PF estimation of 3D ising field. The proposed method is a two-step procedure that applies to the context where multiple 3D ising fields are involved over different objects (eg, brain regions) of different size and topology. First, a small set of reference PFs is accurately estimated using path sampling. Second, the large remaining set of PFs is computed using a temperature-dependent bilinear extrapolation technique. It is shown that our approach is accurate and computationally efficient to account for topological fluctuations of ising fields on regular and irregular graphs. A convincing application to joint detection-estimation of brain activity in functional MRI is also presented.

We present a new algorithm which merges discontinuities in 3-D images of tubular structures presenting undesirable gaps. The application of the proposed method is mainly associated to large 3-D images of microvascular networks. In order to recover the real network topology, we need to fill the gaps between the closest discontinuous vessels. The algorithm presented in this paper aims at achieving this goal. This algorithm is based on the skeletonization of the segmented network followed by a tensor voting method. It permits to merge the most common kinds of discontinuities found in microvascular networks. It is robust, easy to use, and relatively fast. The microvascular network images were obtained using synchrotron tomography imaging at the European Synchrotron Radiation Facility. These images exhibit samples of intracortical networks. Representative results are illustrated.