Search

How can humans remain in control of artificial intelligence (AI)-based systems designed to perform tasks autonomously? Such systems are increasingly ubiquitous, creating benefits - but also undesirable situations where moral responsibility for their actions cannot be properly attributed to any particular person or group. The concept of meaningful human control has been proposed to address responsibility gaps and mitigate them by establishing conditions that enable a proper attribution of responsibility for humans; however, clear requirements for researchers, designers, and engineers are yet inexistent, making the development of AI-based systems that remain under meaningful human control challenging. In this paper, we address the gap between philosophical theory and engineering practice by identifying, through an iterative process of abductive thinking, four actionable properties for AI-based systems under meaningful human control, which we discuss making use of two applications scenarios: automated vehicles and AI-based hiring. First, a system in which humans and AI algorithms interact should have an explicitly defined domain of morally loaded situations within which the system ought to operate. Second, humans and AI agents within the system should have appropriate and mutually compatible representations. Third, responsibility attributed to a human should be commensurate with that human's ability and authority to control the system. Fourth, there should be explicit links between the actions of the AI agents and actions of humans who are aware of their moral responsibility. We argue that these four properties will support practically-minded professionals to take concrete steps toward designing and engineering for AI systems that facilitate meaningful human control., Preprint. Published AI and Ethics (2022): https://doi.org/10.1007/s43681-022-00167-3

Objectives Tenosynovitis (inflammation of the synovial lining of the sheath surrounding tendons) is frequently observed on MRI of early arthritis patients. Since visual assessment of tenosynovitis is a laborious task, we investigated the feasibility of automatic quantification of tenosynovitis on MRI of the wrist in a large cohort of early arthritis patients. Methods For 563 consecutive early arthritis patients (clinically confirmed arthritis ≥ 1 joint, symptoms

Purpose To investigate the feasibility of automatic quantification of bone marrow edema (BME) on MRI of the wrist in patients with early arthritis. Methods For 485 early arthritis patients (clinically confirmed arthritis of one or more joints, symptoms for less than 2 years), MR scans of the wrist were processed in three automatic stages. First, super-resolution reconstruction was applied to fuse coronal and axial scans into a single high-resolution 3D image. Next, the carpal bones were located and delineated using atlas-based segmentation. Finally, the extent of BME within each bone was quantified by identifying image intensity values characteristic of BME by fuzzy clustering and measuring the fraction of voxels with these characteristic intensities within each bone. Correlation with visual BME scores was assessed through Pearson correlation coefficient. Results Pearson correlation between quantitative and visual BME scores across 485 patients was r=0.83, P

Objective MRI-detected inflammation is considered of diagnostic value for rheumatoid arthritis (RA), but its evaluation involves a time-consuming scoring of 61 joint-level features. It is not clear, however, which of these features are specific for RA and whether evaluating a subset of specific features is sufficient to differentiate RA patients. This study aimed to identify a subset of RA-specific features in a case–control setting and validate them in a longitudinal cohort of arthralgia patients. Methods The difference in frequency of MRI-detected inflammation (bone marrow edema, synovitis, and tenosynovitis) between 199 RA patients and 193 controls was studied in 61 features across the wrist, metacarpophalangeal, and metatarsophalangeal joints. A subset of RA-specific features was obtained by applying a cutoff on the frequency difference while maximizing discriminative performance. For validation, this subset was used to predict arthritis development in 225 clinically suspect arthralgia (CSA) patients. Diagnostic performance was compared to a reference method that uses the complete set of 61 features normalized for inflammation levels in age-matched controls. Results Subset of 30 features, mainly (teno)synovitis, was obtained from the case–control setting. Validation in CSA patients yielded an area of 0.69 (95% CI: 0.59–0.78) under the ROC curve and a positive predictive value (PPV) of 31%, compared to 0.68 (95% CI: 0.60–0.77) and 29% PPV of the reference method with 61 features. Conclusion Subset of 30 MRI-detected inflammatory features, dominated by (teno)synovitis, offers a considerable reduction of scoring efforts without compromising accuracy for prediction of arthritis development in CSA patients.

30 pages, 2 figures, pre-print of the paper accepted for publication in the journal Big Data & Society, In the age of big data, companies and governments are increasingly using algorithms to inform hiring decisions, employee management, policing, credit scoring, insurance pricing, and many more aspects of our lives. AI systems can help us make evidence-driven, efficient decisions, but can also confront us with unjustified, discriminatory decisions wrongly assumed to be accurate because they are made automatically and quantitatively. It is becoming evident that these technological developments are consequential to people's fundamental human rights. Despite increasing attention to these urgent challenges in recent years, technical solutions to these complex socio-ethical problems are often developed without empirical study of societal context and the critical input of societal stakeholders who are impacted by the technology. On the other hand, calls for more ethically- and socially-aware AI often fail to provide answers for how to proceed beyond stressing the importance of transparency, explainability, and fairness. Bridging these socio-technical gaps and the deep divide between abstract value language and design requirements is essential to facilitate nuanced, context-dependent design choices that will support moral and social values. In this paper, we bridge this divide through the framework of Design for Values, drawing on methodologies of Value Sensitive Design and Participatory Design to present a roadmap for proactively engaging societal stakeholders to translate fundamental human rights into context-dependent design requirements through a structured, inclusive, and transparent process.

To investigate the feasibility of automatic quantification of bone marrow edema (BME) on MRI of the wrist in patients with early arthritis.For 485 early arthritis patients (clinically confirmed arthritis of one or more joints, symptoms for less than 2 years), MR scans of the wrist were processed in three automatic stages. First, super-resolution reconstruction was applied to fuse coronal and axial scans into a single high-resolution 3D image. Next, the carpal bones were located and delineated using atlas-based segmentation. Finally, the extent of BME within each bone was quantified by identifying image intensity values characteristic of BME by fuzzy clustering and measuring the fraction of voxels with these characteristic intensities within each bone. Correlation with visual BME scores was assessed through Pearson correlation coefficient.Pearson correlation between quantitative and visual BME scores across 485 patients was r=0.83, P0.001.Quantitative measurement of BME on MRI of the wrist has the potential to provide a feasible alternative to visual scoring. Complete automation requires automatic detection and compensation of acquisition artifacts. Magn Reson Med 79:1127-1134, 2018. © 2017 The Authors Magnetic Resonance in Medicine published by Wiley Periodicals, Inc. on behalf of International Society for Magnetic Resonance in Medicine. This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made.

Background: Evaluating inflammatory changes over time on MR images of the spine in patients with suspected axial Spondyloarthritis (axSpA) can be a labor-intensive task, requiring readers to manually search for and perceptually align a set of vertebrae between two scans. The purpose of this study was to assess the feasibility of computer-aided (CA) evaluation of such inflammatory changes in a framework where scans from two time points are fused into a single color-encoded image integrated into an interactive scoring tool. Methods: For 30 patients from the SPondyloArthritis Caught Early (SPACE) cohort (back pain ≥ 3 months, ≤ 2 years, onset < 45 years), baseline and follow-up MR scans acquired 9-12 months apart were fused into a single color-encoded image through locally-rigid image registration to evaluate inflammatory changes in 23 vertebral units (VUs). Scoring was performed by two expert readers on a (-2, 2) scale using an interactive scoring tool. For comparison of direction of change (increase/decrease) indicated by an existing reference, Berlin method scores ((-3, 3) scale) of the same MR scans from a different ongoing study were used. The distributions of VU-level differences between CA readers and between the CA and Berlin methods (sign of change scores) across patients were analyzed descriptively. Patient-level agreement between CA readers was assessed by intraclass correlation coefficient (ICC). Results: Five patients were excluded from evaluation due to failed vertebrae segmentation. Patient-level inter-reader agreement ICC was 0.56 (95% CI: 0.22 to 0.78). Mean VU-level inter-reader differences across 25 patients ranged (-0.04, 0.12) with SD range (0, 0.45). Across all VUs, inter-reader differences ranged (-1, 1) in 573/575 VUs (99.7%). Mean VU-level inter-method differences across patients ranged (-0.04, 0.08) with SD range (0, 0.61). Across all VUs, inter-method differences ranged (-1, 1) in 572/575 VUs (99.5%). Conclusions: Fusion of MR scans of the spine from two time points into a single color-encoded image allows for direct visualization and measurement of inflammatory changes over time in patients with suspected axSpA.

In this paper, we consider a batch learning algorithm for the multilayer neural network with multi-valued neurons (MLMVN) and its soft margins variant (MLMVN-SM). MLMVN is a neural network with a standard feedforward organization based on the multi-valued neuron (MVN). MVN is a neuron with complex-valued weights and inputs/output located on the unit circle. Standard MLMVN has a derivative-free learning algorithm based on the error-correction learning rule. Recently, this algorithm was modified for MLMVN with discrete outputs by using soft margins (MLMVN-SM). This modification improves classification results when MLMVN is used as a classifier. Another recent development in MLMVN is the use of batch acceleration step for MLMVN with a single output neuron. Complex QR-decomposition was used to adjust the output neuron weights for all learning samples simultaneously, while the hidden neuron weights were adjusted in a regular way. In this paper, we merge the soft margins approach with batch learning. We suggest a batch linear least squares (LLS) learning algorithm for MLMVN-SM. We also expand the batch technique to multiple output neurons and hidden neurons. This new learning technique drastically reduces the number of learning iterations and learning time when solving classification problems (compared to MLMVN-SM), while maintaining the classification accuracy of MLMVN-SM.

The use of empirical models for the extraction of optical and physiological properties from reflectance spectra is a relatively new approach, as compared to other techniques such as those based on the diffusion theory and inverse Monte Carlo algorithms. Empirical models are appealing for their ease of implementation and applicability to conditions for which analytical models are limited. Thus far, however, empirical models have been limited to only a handful of specific probe geometries. In this work, the relationship between reflectance and optical property values is explored for a wide range of geometry and tissue conditions. The influence of variation in scattering phase function, and numerical aperture of the optical fibers are investigated and incorporated into the empirical relationships for the first time. Reflectance data used in this work was simulated using an improved Monte Carlo model designed to run on a graphics processing unit (GPU). Improvements include a Modified Henyey-Greenstein and a Mie theory-based phase function in place of the conventional Henyey-Greenstein phase function, and assignment of probe-specific reflectivity conditions to better model the tissue-probe tip interface. These improvements are particularly important for probe geometries with small sourcedetector separations. Probe geometries that offer the most stable relationships between reflectance and optical property values, and hence, the best accuracy and reliability in extraction of physiological properties from tissue, are presented.

The Fourier descriptors paradigm is a well-established approach for affine-invariant characterization of shape contours. In the work presented here, we extend this method to images, and obtain a 2D Fourier representation that is invariant to image rotation. The proposed technique retains phase uniqueness, and therefore structural image information is not lost. Rotation-invariant phase coefficients were used to train a single multi-valued neuron (MVN) to recognize satellite and human face images rotated by a wide range of angles. Experiments yielded 100% and 96.43% classification rate for each data set, respectively. Recognition performance was additionally evaluated under effects of lossy JPEG compression and additive Gaussian noise. Preliminary results show that the derived rotation-invariant features combined with the MVN provide a promising scheme for efficient recognition of rotated images.