Two-step interpretable modeling of ICU-AIs

We present a novel methodology for integrating high resolution longitudinal data with the dynamic prediction capabilities of survival models. The aim is two-fold: to improve the predictive power while maintaining the interpretability of the models. To go beyond the black box paradigm of artificial neural networks, we propose a parsimonious and robust semi-parametric approach (i.e., a landmarking competing risks model) that combines routinely collected low-resolution data with predictive features extracted from a convolutional neural network, that was trained on high resolution time-dependent information. We then use saliency maps to analyze and explain the extra predictive power of this model. To illustrate our methodology, we focus on healthcare-associated infections in patients admitted to an intensive care unit.

SaRF: Saliency regularized feature learning improves MRI sequence classification

BACKGROUND AND OBJECTIVE

Deep learning based medical image analysis technologies have the potential to greatly improve the workflow of neuro-radiologists dealing routinely with multi-sequence MRI. However, an essential step for current deep learning systems employing multi-sequence MRI is to ensure that their sequence type is correctly assigned. This requirement is not easily satisfied in clinical practice and is subjected to protocol and human-prone errors. Although deep learning models are promising for image-based sequence classification, robustness, and reliability issues limit their application to clinical practice.

METHODS

In this paper, we propose a novel method that uses saliency information to guide the learning of features for sequence classification. The method uses two self-supervised loss terms to first enhance the distinctiveness among class-specific saliency maps and, secondly, to promote similarity between class-specific saliency maps and learned deep features.

RESULTS

On a cohort of 2100 patient cases comprising six different MR sequences per case, our method shows an improvement in mean accuracy by 4.4% (from 0.935 to 0.976), mean AUC by 1.2% (from 0.9851 to 0.9968), and mean F1 score by 20.5% (from 0.767 to 0.924). Furthermore, based on feedback from an expert neuroradiologist, we show that the proposed approach improves the interpretability of trained models as well as their calibration with reduced expected calibration error (by 30.8%, from 0.065 to 0.045). The code will be made publicly available.

CONCLUSIONS

In this paper, the proposed method shows an improvement in accuracy, AUC, and F1 score, as well as improved calibration and interpretability of resulting saliency maps.

Sensory salience processing moderates attenuated gazes on faces in autism spectrum disorder: a case-control study

BACKGROUND

Attenuated social attention is a key marker of autism spectrum disorder (ASD). Recent neuroimaging findings also emphasize an altered processing of sensory salience in ASD. The locus coeruleus-norepinephrine system (LC-NE) has been established as a modulator of this sensory salience processing (SSP). We tested the hypothesis that altered LC-NE functioning contributes to different SSP and results in diverging social attention in ASD.

METHODS

We analyzed the baseline eye-tracking data of the EU-AIMS Longitudinal European Autism Project (LEAP) for subgroups of autistic participants (n = 166, age = 6-30 years, IQ = 61-138, gender [female/male] = 41/125) or neurotypical development (TD; n = 166, age = 6-30 years, IQ = 63-138, gender [female/male] = 49/117) that were matched for demographic variables and data quality. Participants watched brief movie scenes (k = 85) depicting humans in social situations (human) or without humans (non-human). SSP was estimated by gazes on physical and motion salience and a corresponding pupillary response that indexes phasic activity of the LC-NE. Social attention is estimated by gazes on faces via manual areas of interest definition. SSP is compared between groups and related to social attention by linear mixed models that consider temporal dynamics within scenes. Models are controlled for comorbid psychopathology, gaze behavior, and luminance.

RESULTS

We found no group differences in gazes on salience, whereas pupillary responses were associated with altered gazes on physical and motion salience. In ASD compared to TD, we observed pupillary responses that were higher for non-human scenes and lower for human scenes. In ASD, we observed lower gazes on faces across the duration of the scenes. Crucially, this different social attention was influenced by gazes on physical salience and moderated by pupillary responses.

LIMITATIONS

The naturalistic study design precluded experimental manipulations and stimulus control, while effect sizes were small to moderate. Covariate effects of age and IQ indicate that the findings differ between age and developmental subgroups.

CONCLUSIONS

Pupillary responses as a proxy of LC-NE phasic activity during visual attention are suggested to modulate sensory salience processing and contribute to attenuated social attention in ASD.

Visualising and quantifying relevant parkinsonian gait patterns using 3D convolutional network

Parkinson's disease (PD) lacks a definitive diagnosis, with the observation of motion patterns being the main method of characterizing disease progression and planning patient treatments. Among PD observations, gait motion patterns, such as step length, flexed posture, and bradykinesia, support the characterization of disease progression. However, this analysis is usually performed with marker-based protocols, which affect the gait and localized segment patterns during locomotion. This work introduces a 3D convolutional gait representation for automatic PD classification that identifies the spatio-temporal patterns used for classification. This approach allows us to obtain an explainable model that classifies markerless sequences and describes the main learned spatio-temporal regions associated with abnormal patterns in a particular video. Initially, a spatio-temporal convolutional network is trained from a set of raw videos and optical flow fields. Then, a PD prediction is obtained from the motion patterns learned by the trained model. Finally, saliency maps, which highlight abnormal motion patterns, are obtained by retro-propagating the output prediction up to the input volume through two different stages: an embedded back-tracking and a pseudo-deconvolution process. From a total of 176 videos from 22 patients, the resulting salient maps highlight lower limb patterns possibly related to step length and speed. In control subjects, the saliency maps highlight the head and trunk posture. The proposed approach achieved an average accuracy score of 94.89%.

COVID-19 pulmonary consolidations detection in chest X-ray using progressive resizing and transfer learning techniques

A viral outbreak with a lower respiratory tract febrile illness causes pulmonary syndrome named COVID-19. Pulmonary consolidations developed in the lungs of the patients are imperative factors during prognosis and diagnosis. Existing Deep Learning techniques demonstrate promising results in analyzing X-ray images when employed with Transfer Learning. However, Transfer Learning has its inherent limitations, which can be prevaricated by employing the Progressive Resizing technique. The Progressive Resizing technique reuses old computations while learning new ones in Convolution Neural Networks (CNN), enabling it to incorporate prior knowledge of the feature hierarchy. The proposed classification model can classify pulmonary consolidation into normal, pneumonia, and SARS-CoV-2 classes by analyzing X-rays images. The method exhibits substantial enhancement in classification results when the Transfer Learning technique is applied in consultation with the Progressive Resizing technique on EfficientNet CNN. The customized VGG-19 model attained benchmark scores in all evaluation criteria over the baseline VGG-19 model. GradCam based feature interpretation, coupled with X-ray visual analysis, facilitates improved assimilation of the scores. The model highlights its strength to assist medical experts in the COVID-19 identification during the prognosis and subsequently for diagnosis. Clinical implications exist in peripheral and remotely located health centers with the paucity of trained human resources to interpret radiological investigations' findings.

Neural foundations and functional specificity of number representations

Number is a complex category, as with the word "number" we may refer to different entities. First, it is a perceptual property that characterizes any set of individual items, namely its cardinality. The ability to extract the (approximate) cardinality of sets is almost universal in the animal domain and present in humans since birth. In primates, posterior parietal cortex seems to be a crucial site for this ability, even if the degree of selectivity of numerical representations in parietal cortex reported to date appears much lower compared to that of other semantic categories in the ventral stream. Number can also be intended as a mathematical object, which we humans use to count, measure, and order: a (verbal or visual) symbol that stands for the cardinality of a set, the intensity of a continuous quantity or the position of an item on a list. Evidence points to a convergence towards parietal cortex for the semantic coding of numerical symbols and to the bilateral occipitotemporal cortex for the shape coding of Arabic digits and other number symbols.