Creating visual explanations improves learning.
Ontology highlight
ABSTRACT:
SUBMITTER: Bobek E
PROVIDER: S-EPMC5256450 | biostudies-literature |
REPOSITORIES: biostudies-literature
Similar Datasets
Project description:It has been shown that identical deep learning (DL) architectures will produce distinct explanations when trained with different hyperparameters that are orthogonal to the task (e.g. random seed, training set order). In domains such as healthcare and finance, where transparency and explainability is paramount, this can be a significant barrier to DL adoption. In this study we present a further analysis of explanation (in)consistency on 6 tabular datasets/tasks, with a focus on Electronic Health Records data. We propose a novel deep learning ensemble architecture that trains its sub-models to produce consistent explanations, improving explanation consistency by as much as 315% (e.g. from 0.02433 to 0.1011 on MIMIC-IV), and on average by 124% (e.g. from 0.12282 to 0.4450 on the BCW dataset). We evaluate the effectiveness of our proposed technique and discuss the implications our results have for both industrial applications of DL and explainability as well as future methodological work.
| S-EPMC9674624 | biostudies-literature
Project description:By emulating biological features in brain, Spiking Neural Networks (SNNs) offer an energy-efficient alternative to conventional deep learning. To make SNNs ubiquitous, a 'visual explanation' technique for analysing and explaining the internal spike behavior of such temporal deep SNNs is crucial. Explaining SNNs visually will make the network more transparent giving the end-user a tool to understand how SNNs make temporal predictions and why they make a certain decision. In this paper, we propose a bio-plausible visual explanation tool for SNNs, called Spike Activation Map (SAM). SAM yields a heatmap (i.e., localization map) corresponding to each time-step of input data by highlighting neurons with short inter-spike interval activity. Interestingly, without the use of gradients and ground truth, SAM produces a temporal localization map highlighting the region of interest in an image attributed to an SNN's prediction at each time-step. Overall, SAM outsets the beginning of a new research area 'explainable neuromorphic computing' that will ultimately allow end-users to establish appropriate trust in predictions from SNNs.
| S-EPMC8463578 | biostudies-literature
Project description:Real-time rendering of closed-loop visual environments is important for next-generation understanding of brain function and behaviour, but is often prohibitively difficult for non-experts to implement and is limited to few laboratories worldwide. We developed BonVision as an easy-to-use open-source software for the display of virtual or augmented reality, as well as standard visual stimuli. BonVision has been tested on humans and mice, and is capable of supporting new experimental designs in other animal models of vision. As the architecture is based on the open-source Bonsai graphical programming language, BonVision benefits from native integration with experimental hardware. BonVision therefore enables easy implementation of closed-loop experiments, including real-time interaction with deep neural networks, and communication with behavioural and physiological measurement and manipulation devices.
| S-EPMC8104957 | biostudies-literature
Project description:Artificial intelligence (AI) systems can now reliably assess surgeon skills through videos of intraoperative surgical activity. With such systems informing future high-stakes decisions such as whether to credential surgeons and grant them the privilege to operate on patients, it is critical that they treat all surgeons fairly. However, it remains an open question whether surgical AI systems exhibit bias against surgeon sub-cohorts, and, if so, whether such bias can be mitigated. Here, we examine and mitigate the bias exhibited by a family of surgical AI systems-SAIS-deployed on videos of robotic surgeries from three geographically-diverse hospitals (USA and EU). We show that SAIS exhibits an underskilling bias, erroneously downgrading surgical performance, and an overskilling bias, erroneously upgrading surgical performance, at different rates across surgeon sub-cohorts. To mitigate such bias, we leverage a strategy -TWIX-which teaches an AI system to provide a visual explanation for its skill assessment that otherwise would have been provided by human experts. We show that whereas baseline strategies inconsistently mitigate algorithmic bias, TWIX can effectively mitigate the underskilling and overskilling bias while simultaneously improving the performance of these AI systems across hospitals. We discovered that these findings carry over to the training environment where we assess medical students' skills today. Our study is a critical prerequisite to the eventual implementation of AI-augmented global surgeon credentialing programs, ensuring that all surgeons are treated fairly.
| S-EPMC10063676 | biostudies-literature
Project description:Deep neural networks are complex machine learning models that have shown promising results in analyzing high-dimensional data such as those collected from medical examinations. Such models have the potential to provide fast and accurate medical diagnoses. However, the high complexity makes deep neural networks and their predictions difficult to understand. Providing model explanations can be a way of increasing the understanding of "black box" models and building trust. In this work, we applied transfer learning to develop a deep neural network to predict sex from electrocardiograms. Using the visual explanation method Grad-CAM, heat maps were generated from the model in order to understand how it makes predictions. To evaluate the usefulness of the heat maps and determine if the heat maps identified electrocardiogram features that could be recognized to discriminate sex, medical doctors provided feedback. Based on the feedback, we concluded that, in our setting, this mode of explainable artificial intelligence does not provide meaningful information to medical doctors and is not useful in the clinic. Our results indicate that improved explanation techniques that are tailored to medical data should be developed before deep neural networks can be applied in the clinic for diagnostic purposes.
| S-EPMC10378376 | biostudies-literature
Project description:Machine learning methods are widely used within the medical field. However, the reliability and efficacy of these models is difficult to assess, making it difficult for researchers to identify which machine-learning model to apply to their dataset. We assessed whether variance calculations of model metrics (e.g., AUROC, Sensitivity, Specificity) through bootstrap simulation and SHapely Additive exPlanations (SHAP) could increase model transparency and improve model selection. Data from the England National Health Services Heart Disease Prediction Cohort was used. After comparison of model metrics for XGBoost, Random Forest, Artificial Neural Network, and Adaptive Boosting, XGBoost was used as the machine-learning model of choice in this study. Boost-strap simulation (N = 10,000) was used to empirically derive the distribution of model metrics and covariate Gain statistics. SHapely Additive exPlanations (SHAP) to provide explanations to machine-learning output and simulation to evaluate the variance of model accuracy metrics. For the XGBoost modeling method, we observed (through 10,000 completed simulations) that the AUROC ranged from 0.771 to 0.947, a difference of 0.176, the balanced accuracy ranged from 0.688 to 0.894, a 0.205 difference, the sensitivity ranged from 0.632 to 0.939, a 0.307 difference, and the specificity ranged from 0.595 to 0.944, a 0.394 difference. Among 10,000 simulations completed, we observed that the gain for Angina ranged from 0.225 to 0.456, a difference of 0.231, for Cholesterol ranged from 0.148 to 0.326, a difference of 0.178, for maximum heart rate (MaxHR) ranged from 0.081 to 0.200, a range of 0.119, and for Age ranged from 0.059 to 0.157, difference of 0.098. Use of simulations to empirically evaluate the variability of model metrics and explanatory algorithms to observe if covariates match the literature are necessary for increased transparency, reliability, and utility of machine learning methods. These variance statistics, combined with model accuracy statistics can help researchers identify the best model for a given dataset.
| S-EPMC9949629 | biostudies-literature
Project description:BackgroundInterest in value-based healthcare, generally defined as providing better care at lower cost, has grown worldwide, and learning health systems (LHSs) have been proposed as a key strategy for improving value in healthcare. LHSs are emerging around the world and aim to leverage advancements in science, technology and practice to improve health system performance at lower cost. However, there remains much uncertainty around the implementation of LHSs and the distinctive features of these systems. This paper presents a conceptual framework that has been developed in Canada to support the implementation of value-creating LHSs.MethodsThe framework was developed by an interdisciplinary team at the Institut national d'excellence en santé et en services sociaux (INESSS). It was informed by a scoping review of the scientific and grey literature on LHSs, regular team discussions over a 14-month period, and consultations with Canadian and international experts.ResultsThe framework describes four elements that characterise LHSs, namely (1) core values, (2) pillars and accelerators, (3) processes and (4) outcomes. LHSs embody certain core values, including an emphasis on participatory leadership, inclusiveness, scientific rigour and person-centredness. In addition, values such as equity and solidarity should also guide LHSs and are particularly relevant in countries like Canada. LHS pillars are the infrastructure and resources supporting the LHS, whereas accelerators are those specific structures that enable more rapid learning and improvement. For LHSs to create value, such infrastructures must not only exist within the ecosystem but also be connected and aligned with the LHSs' strategic goals. These pillars support the execution, routinisation and acceleration of learning cycles, which are the fundamental processes of LHSs. The main outcome sought by executing learning cycles is the creation of value, which we define as the striking of a more optimal balance of impacts on patient and provider experience, population health and health system costs.ConclusionsOur framework illustrates how the distinctive structures, processes and outcomes of LHSs tie together with the aim of optimising health system performance and delivering greater value in health systems.
| S-EPMC6688264 | biostudies-literature
Project description:It has been argued that novel compared to familiar stimuli are preferentially encoded into memory. Nevertheless, treating novelty as a categorical variable in experimental research is considered simplistic. We highlight the dimensional aspect of novelty and propose an experimental design that manipulates novelty continuously. We created the Graded Novelty Encoding Task (GNET), in which the difference between stimuli (i.e. novelty) is parametrically manipulated, paving the way for quantitative models of novelty processing. We designed an algorithm which generates visual stimuli by placing colored shapes in a grid. During the familiarization phase of the task, we repeatedly presented five pictures to the participants. In a subsequent incidental learning phase, participants were asked to differentiate between the "familiars" and novel images that varied in the degree of difference to the familiarized pictures (i.e. novelty). Finally, participants completed a surprise recognition memory test, where the novel stimuli from the previous phase were interspersed with distractors with similar difference characteristics. We numerically expressed the differences between the stimuli to compute a dimensional indicator of novelty and assessed whether it predicted recognition memory performance. Based on previous studies showing the beneficial effect of novelty on memory formation, we hypothesized that the more novel a given picture was, the better subsequent recognition performance participants would demonstrate. Our hypothesis was confirmed: recognition performance was higher for more novel stimuli. The GNET captures the continuous nature of novelty, and it may be useful in future studies that examine the behavioral and neurocognitive aspects of novelty processing.
| S-EPMC10250520 | biostudies-literature
Project description:BackgroundResearch has found that human associated microbial communities play a role in homeostasis and the disruption of these communities may be important in an array of medical conditions. However outside of the human body many of these communities remain poorly studied. The Metagenomics and Metadesign of the Subways and Urban Biomes (MetaSUB) International Consortium is characterizing the microbiomes of urban environments with the aim to improve design of mass transit systems. As part of the CAMDA 2018 MetaSUB Forensics Challenge 311 city microbiome samples were provided to create urban microbial fingerprints, as well as a further 3 mystery datasets for validation.ResultsMetaSUB samples were clustered using t-SNE in an unsupervised fashion to almost discrete groups, which upon inspection represented city of origin. Based on this clustering, geographically close metropolitan areas appear to display similar microbial profiles such as those of Auckland and Hamilton. Mystery unlabeled samples were provided part of the challenge. A random forest classifier built on the initial dataset of 311 samples was capable of correctly classifying 83.3% of the mystery samples to their city of origin. Random Forest analyses also identified features with the highest discriminatory power, ranking bacterial species such as Campylobacter jejuni and Staphylococcus argenteus as highly predictive of city of origin. The surface from which the sample was collected displayed little detectable impact on the microbial profiles in the data generated here. The proportion of reads classified per sample varied greatly and so de-novo assembly was applied to recover genomic fragments representing organisms not captured in reference databases.ConclusionsCurrent methods can differentiate urban microbiome profiles from each other with relative ease. De-novo assembly indicated that the MetaSUB metagenomic data contains adequate depth to recover metagenomic assembled genomes and that current databases are not sufficient to fully characterize urban microbiomes. Profiles found here indicate there may be a relationship between geographical distance between areas and the urban microbiome composition although this will need further research. The impact of these different profiles on public health is currently unknown but the MetaSUB consortium is uniquely suited to evaluate these and provide a roadmap for the inclusion of urban microbiome information for city planning and public health policy.ReviewersThis article was reviewed by Dimitar Vassilev, Eran Elhaik and Chengsheng Zhu.
| S-EPMC6697990 | biostudies-literature
Project description:Visual-to-auditory sensory substitution devices (SSDs) convey visual information via sound, with the primary goal of making visual information accessible to blind and visually impaired individuals. We developed the EyeMusic SSD, which transforms shape, location, and color information into musical notes. We tested the "visual" acuity of 23 individuals (13 blind and 10 blindfolded sighted) on the Snellen tumbling-E test, with the EyeMusic. Participants were asked to determine the orientation of the letter "E." The test was repeated twice: in one test, the letter "E" was drawn with a single color (white), and in the other test, with two colors (red and white). In the latter case, the vertical line in the letter, when upright, was drawn in red, with the three horizontal lines drawn in white. We found no significant differences in performance between the blind and the sighted groups. We found a significant effect of the added color on the "visual" acuity. The highest acuity participants reached in the monochromatic test was 20/800, whereas with the added color, acuity doubled to 20/400. We conclude that color improves "visual" acuity via sound.
| S-EPMC4227506 | biostudies-literature