Computing linkage disequilibrium aware genome embeddings using autoencoders.

MOTIVATION: The completion of the genome has paved the way for genome-wide association studies (GWAS), which explained certain proportions of heritability. GWAS are not optimally suited to detect non-linear effects in disease risk, possibly hidden in non-additive interactions (epistasis). Alternative methods for epistasis detection using, e.g. deep neural networks (DNNs) are currently under active development. However, DNNs are constrained by finite computational resources, which can be rapidly depleted due to increasing complexity with the sheer size of the genome. Besides, the curse of dimensionality complicates the task of capturing meaningful genetic patterns for DNNs; therefore necessitates dimensionality reduction. RESULTS: We propose a method to compress single nucleotide polymorphism (SNP) data, while leveraging the linkage disequilibrium (LD) structure and preserving potential epistasis. This method involves clustering correlated SNPs into haplotype blocks and training per-block autoencoders to learn a compressed representation of the block's genetic content. We provide an adjustable autoencoder design to accommodate diverse blocks and bypass extensive hyperparameter tuning. We applied this method to genotyping data from Project MinE, and achieved 99% average test reconstruction accuracy-i.e. minimal information loss-while compressing the input to nearly 10% of the original size. We demonstrate that haplotype-block based autoencoders outperform linear Principal Component Analysis (PCA) by approximately 3% chromosome-wide accuracy of reconstructed variants. To the extent of our knowledge, our approach is the first to simultaneously leverage haplotype structure and DNNs for dimensionality reduction of genetic data. AVAILABILITY AND IMPLEMENTATION: Data are available for academic use through Project MinE at https://www.projectmine.com/research/data-sharing/, contingent upon terms and requirements specified by the source studies. Code is available at https://github.com/gizem-tas/haploblock-autoencoders.

Predicting vasovagal reactions to a virtual blood donation using facial image analysis.

BACKGROUND: People with needle fear experience not only anxiety and stress but also vasovagal reactions (VVR), including nausea, dizziness, sweating, pallor changes, or even fainting. However, the mechanism behind needle fear and the VVR response are not yet well understood. The aim of our study was to explore whether fluctuations in facial temperature in several facial regions are related to the level of experienced vasovagal reactions, in a simulated blood donation. STUDY DESIGN AND METHODS: We recruited 45 students at Tilburg University and filmed them throughout a virtual blood donation procedure using an Infrared Thermal Imaging (ITI) camera. Participants reported their fear of needles and level of experienced vasovagal reactions. ITI data pre-processing was completed on each video frame by detecting facial landmarks and image alignment before extracting the mean temperature from the six regions of interest. RESULTS: Temperatures of the chin and left and right cheek areas increased during the virtual blood donation. Mixed-effects linear regression showed a significant association between self-reported vasovagal reactions and temperature fluctuations in the area below the nose. DISCUSSION: Our results suggest that the area below the nose may be an interesting target for measuring vasovagal reactions using video imaging techniques. This is the first in a line of studies, which assess whether it is possible to automatically detect levels of fear and vasovagal reactions using facial imaging, from which the development of e-health solutions and interventions can benefit.

Automatic detection of confusion in elderly users of a web-based health instruction video.

BACKGROUND: Because of cognitive limitations and lower health literacy, many elderly patients have difficulty understanding verbal medical instructions. Automatic detection of facial movements provides a nonintrusive basis for building technological tools supporting confusion detection in healthcare delivery applications on the Internet. MATERIALS AND METHODS: Twenty-four elderly participants (70-90 years old) were recorded while watching Web-based health instruction videos involving easy and complex medical terminology. Relevant fragments of the participants' facial expressions were rated by 40 medical students for perceived level of confusion and analyzed with automatic software for facial movement recognition. RESULTS: A computer classification of the automatically detected facial features performed more accurately and with a higher sensitivity than the human observers (automatic detection and classification, 64% accuracy, 0.64 sensitivity; human observers, 41% accuracy, 0.43 sensitivity). A drill-down analysis of cues to confusion indicated the importance of the eye and eyebrow region. CONCLUSIONS: Confusion caused by misunderstanding of medical terminology is signaled by facial cues that can be automatically detected with currently available facial expression detection technology. The findings are relevant for the development of Web-based services for healthcare consumers.

Synthetic images aid the recognition of human-made art forgeries.

Previous research has shown that Artificial Intelligence is capable of distinguishing between authentic paintings by a given artist and human-made forgeries with remarkable accuracy, provided sufficient training. However, with the limited amount of existing known forgeries, augmentation methods for forgery detection are highly desirable. In this work, we examine the potential of incorporating synthetic artworks into training datasets to enhance the performance of forgery detection. Our investigation focuses on paintings by Vincent van Gogh, for which we release the first dataset specialized for forgery detection. To reinforce our results, we conduct the same analyses on the artists Amedeo Modigliani and Raphael. We train a classifier to distinguish original artworks from forgeries. For this, we use human-made forgeries and imitations in the style of well-known artists and augment our training sets with images in a similar style generated by Stable Diffusion and StyleGAN. We find that the additional synthetic forgeries consistently improve the detection of human-made forgeries. In addition, we find that, in line with previous research, the inclusion of synthetic forgeries in the training also enables the detection of AI-generated forgeries, especially if created using a similar generator.

Cognitive functioning in untreated glioma patients: The limited predictive value of clinical variables.

BACKGROUND: Previous research identified many clinical variables that are significantly related to cognitive functioning before surgery. It is not clear whether such variables enable accurate prediction for individual patients' cognitive functioning because statistical significance does not guarantee predictive value. Previous studies did not test how well cognitive functioning can be predicted for (yet) untested patients. Furthermore, previous research is limited in that only linear or rank-based methods with small numbers of variables were used. METHODS: We used various machine learning models to predict preoperative cognitive functioning for 340 patients with glioma across 18 outcome measures. Predictions were made using a comprehensive set of clinical variables as identified from the literature. Model performances and optimized hyperparameters were interpreted. Moreover, Shapley additive explanations were calculated to determine variable importance and explore interaction effects. RESULTS: Best-performing models generally demonstrated above-random performance. Performance, however, was unreliable for 14 out of 18 outcome measures with predictions worse than baseline models for a substantial number of train-test splits. Best-performing models were relatively simple and used most variables for prediction while not relying strongly on any variable. CONCLUSIONS: Preoperative cognitive functioning could not be reliably predicted across cognitive tests using the comprehensive set of clinical variables included in the current study. Our results show that a holistic view of an individual patient likely is necessary to explain differences in cognitive functioning. Moreover, they emphasize the need to collect larger cross-center and multimodal data sets.

Development and Validation of a Convolutional Neural Network for Automated Detection of Scaphoid Fractures on Conventional Radiographs.

PURPOSE: To compare the performance of a convolutional neural network (CNN) to that of 11 radiologists in detecting scaphoid bone fractures on conventional radiographs of the hand, wrist, and scaphoid. MATERIALS AND METHODS: At two hospitals (hospitals A and B), three datasets consisting of conventional hand, wrist, and scaphoid radiographs were retrospectively retrieved: a dataset of 1039 radiographs (775 patients [mean age, 48 years ± 23 {standard deviation}; 505 female patients], period: 2017-2019, hospitals A and B) for developing a scaphoid segmentation CNN, a dataset of 3000 radiographs (1846 patients [mean age, 42 years ± 22; 937 female patients], period: 2003-2019, hospital B) for developing a scaphoid fracture detection CNN, and a dataset of 190 radiographs (190 patients [mean age, 43 years ± 20; 77 female patients], period: 2011-2020, hospital A) for testing the complete fracture detection system. Both CNNs were applied consecutively: The segmentation CNN localized the scaphoid and then passed the relevant region to the detection CNN for fracture detection. In an observer study, the performance of the system was compared with that of 11 radiologists. Evaluation metrics included the Dice similarity coefficient (DSC), Hausdorff distance (HD), sensitivity, specificity, positive predictive value (PPV), and area under the receiver operating characteristic curve (AUC). RESULTS: The segmentation CNN achieved a DSC of 97.4% ± 1.4 with an HD of 1.31 mm ± 1.03. The detection CNN had sensitivity of 78% (95% CI: 70, 86), specificity of 84% (95% CI: 77, 92), PPV of 83% (95% CI: 77, 90), and AUC of 0.87 (95% CI: 0.81, 0.91). There was no difference between the AUC of the CNN and that of the radiologists (0.87 [95% CI: 0.81, 0.91] vs 0.83 [radiologist range: 0.79-0.85]; P = .09). CONCLUSION: The developed CNN achieved radiologist-level performance in detecting scaphoid bone fractures on conventional radiographs of the hand, wrist, and scaphoid.Keywords: Convolutional Neural Network (CNN), Deep Learning Algorithms, Machine Learning Algorithms, Feature Detection-Vision-Application Domain, Computer-Aided DiagnosisSee also the commentary by Li and Torriani in this issue.Supplemental material is available for this article.©RSNA, 2021.

Combining structural connectivity and response latencies to model the structure of the visual system.

Several approaches exist to ascertain the connectivity of the brain, and these approaches lead to markedly different topologies, often incompatible with each other. Specifically, recent single-cell recording results seem incompatible with current structural connectivity models. We present a novel method that combines anatomical and temporal constraints to generate biologically plausible connectivity patterns of the visual system of the macaque monkey. Our method takes structural connectivity data from the CoCoMac database and recent single-cell recording data as input and employs an optimization technique to arrive at a new connectivity pattern of the visual system that is in agreement with both types of experimental data. The new connectivity pattern yields a revised model that has fewer levels than current models. In addition, it introduces subcortical-cortical connections. We show that these connections are essential for explaining latency data, are consistent with our current knowledge of the structural connectivity of the visual system, and might explain recent functional imaging results in humans. Furthermore we show that the revised model is not underconstrained like previous models and can be extended to include newer data and other kinds of data. We conclude that the revised model of the connectivity of the visual system reflects current knowledge on the structure and function of the visual system and addresses some of the limitations of previous models.

Modeling recognition memory using the similarity structure of natural input.

The natural input memory (NIM) model is a new model for recognition memory that operates on natural visual input. A biologically informed perceptual preprocessing method takes local samples (eye fixations) from a natural image and translates these into a feature-vector representation. During recognition, the model compares incoming preprocessed natural input to stored representations. By complementing the recognition memory process with a perceptual front end, the NIM model is able to make predictions about memorability based directly on individual natural stimuli. We demonstrate that the NIM model is able to simulate experimentally obtained similarity ratings and recognition memory for individual stimuli (i.e., face images).

Comparing a Perceptual and an Automated Vision-Based Method for Lie Detection in Younger Children.

The present study investigates how easily it can be detected whether a child is being truthful or not in a game situation, and it explores the cue validity of bodily movements for such type of classification. To achieve this, we introduce an innovative methodology - the combination of perception studies (in which eye-tracking technology is being used) and automated movement analysis. Film fragments from truthful and deceptive children were shown to human judges who were given the task to decide whether the recorded child was being truthful or not. Results reveal that judges are able to accurately distinguish truthful clips from lying clips in both perception studies. Even though the automated movement analysis for overall and specific body regions did not yield significant results between the experimental conditions, we did find a positive correlation between the amount of movement in a child and the perception of lies, i.e., the more movement the children exhibited during a clip, the higher the chance that the clip was perceived as a lie. The eye-tracking study revealed that, even when there is movement happening in different body regions, judges tend to focus their attention mainly on the face region. This is the first study that compares a perceptual and an automated method for the detection of deceptive behavior in children whose data have been elicited through an ecologically valid paradigm.

Auditory perception bias in speech imitation.

In an experimental study, we explored the role of auditory perception bias in vocal pitch imitation. Psychoacoustic tasks involving a missing fundamental indicate that some listeners are attuned to the relationship between all the higher harmonics present in the signal, which supports their perception of the fundamental frequency (the primary acoustic correlate of pitch). Other listeners focus on the lowest harmonic constituents of the complex sound signal which may hamper the perception of the fundamental. These two listener types are referred to as fundamental and spectral listeners, respectively. We hypothesized that the individual differences in speakers' capacity to imitate F 0 found in earlier studies, may at least partly be due to the capacity to extract information about F 0 from the speech signal. Participants' auditory perception bias was determined with a standard missing fundamental perceptual test. Subsequently, speech data were collected in a shadowing task with two conditions, one with a full speech signal and one with high-pass filtered speech above 300 Hz. The results showed that perception bias toward fundamental frequency was related to the degree of F 0 imitation. The effect was stronger in the condition with high-pass filtered speech. The experimental outcomes suggest advantages for fundamental listeners in communicative situations where F 0 imitation is used as a behavioral cue. Future research needs to determine to what extent auditory perception bias may be related to other individual properties known to improve imitation, such as phonetic talent.

Development and external validation of a predictive model for pathological complete response of rectal cancer patients including sequential PET-CT imaging.

PURPOSE: To develop and validate an accurate predictive model and a nomogram for pathologic complete response (pCR) after chemoradiotherapy (CRT) for rectal cancer based on clinical and sequential PET-CT data. Accurate prediction could enable more individualised surgical approaches, including less extensive resection or even a wait-and-see policy. METHODS AND MATERIALS: Population based databases from 953 patients were collected from four different institutes and divided into three groups: clinical factors (training: 677 patients, validation: 85 patients), pre-CRT PET-CT (training: 114 patients, validation: 37 patients) and post-CRT PET-CT (training: 107 patients, validation: 55 patients). A pCR was defined as ypT0N0 reported by pathology after surgery. The data were analysed using a linear multivariate classification model (support vector machine), and the model's performance was evaluated using the area under the curve (AUC) of the receiver operating characteristic (ROC) curve. RESULTS: The occurrence rate of pCR in the datasets was between 15% and 31%. The model based on clinical variables (AUC(train)=0.61±0.03, AUC(validation)=0.69±0.08) resulted in the following predictors: cT- and cN-stage and tumour length. Addition of pre-CRT PET data did not result in a significantly higher performance (AUC(train)=0.68±0.08, AUC(validation)=0.68±0.10) and revealed maximal radioactive isotope uptake (SUV(max)) and tumour location as extra predictors. The best model achieved was based on the addition of post-CRT PET-data (AUC(train)=0.83±0.05, AUC(validation)=0.86±0.05) and included the following predictors: tumour length, post-CRT SUV(max) and relative change of SUV(max). This model performed significantly better than the clinical model (p(train)<0.001, p(validation)=0.056). CONCLUSIONS: The model and the nomogram developed based on clinical and sequential PET-CT data can accurately predict pCR, and can be used as a decision support tool for surgery after prospective validation.

Non-parametric algorithmic generation of neuronal morphologies.

Generation algorithms allow for the generation of Virtual Neurons (VNs) from a small set of morphological properties. The set describes the morphological properties of real neurons in terms of statistical descriptors such as the number of branches and segment lengths (among others). The majority of reconstruction algorithms use the observed properties to estimate the parameters of a priori fixed probability distributions in order to construct statistical descriptors that fit well with the observed data. In this article, we present a non-parametric generation algorithm based on kernel density estimators (KDEs). The new algorithm is called KDE-NEURON: and has three advantages over parametric reconstruction algorithms: (1) no a priori specifications about the distributions underlying the real data, (2) peculiarities in the biological data will be reflected in the VNs, and (3) ability to reconstruct different cell types. We experimentally generated motor neurons and granule cells, and statistically validated the obtained results. Moreover, we assessed the quality of the prototype data set and observed that our generated neurons are as good as the prototype data in terms of the used statistical descriptors. The opportunities and limitations of data-driven algorithmic reconstruction of neurons are discussed.

SCAN: A Scalable Model of Attentional Selection.

This paper describes the SCAN (Signal Channelling Attentional Network) model, a scalable neural network model for attentional scanning. The building block of SCAN is a gating lattice, a sparsely-connected neural network defined as a special case of the Ising lattice from statistical mechanics. The process of spatial selection through covert attention is interpreted as a biological solution to the problem of translation-invariant pattern processing. In SCAN, a sequence of pattern translations combines active selection with translation-invariant processing. Selected patterns are channelled through a gating network, formed by a hierarchical fractal structure of gating lattices, and mapped onto an output window. We show how the incorporation of an expectation-generating classifier network (e.g. Carpenter and Grossberg's ART network) into SCAN allows attentional selection to be driven by expectation. Simulation studies show the SCAN model to be capable of attending and identifying object patterns that are part of a realistically sized natural image. Copyright 1997 Elsevier Science Ltd.