The goal of the present study was to further test the hypothesis that objects are important units of saccade targeting and, by inference, attentional selection in real-world scene perception. To this end, we investigated where people fixate within objects embedded in natural scenes. Previously, we reported a preferred viewing location (PVL) close to the center of objects (Nuthmann & Henderson, 2010). Here, we qualify this basic finding by showing that the PVL is affected by object size and the distance between the object and the previous fixation (i.e., launch site distance). Moreover, we examined how within-object fixation position affected subsequent eye-movement behavior on the object. Unexpectedly, there was no refixation optimal viewing position (OVP) effect for objects in scenes. Where viewers initially placed their eyes on an object did not affect the likelihood of refixating that object, suggesting that some refixations on objects in scenes are made for reasons other than insufficient visual information. A fixation-duration inverted-optimal viewing (IOVP) effect was found for large objects: Fixations located at object center were longer than those falling near the edges of an object. Collectively, these findings lend further support to the notion of object-based saccade targeting in scenes.

We present a semantically interpretable system for automated ICD coding of clinical text documents. Our contribution is an ontological attention mechanism which matches the structure of the ICD ontology, in which shared attention vectors are learned at each level of the hierarchy, and combined into label-dependent ensembles. Analysis of the attention heads shows that shared concepts are learned by the lowest common denominator node. This allows child nodes to focus on the differentiating concepts, leading to efficient learning and memory usage. Visualisation of the multilevel attention on the original text allows explanation of the code predictions according to the semantics of the ICD ontology. On the MIMIC-III dataset we achieve a 2.7% absolute (11% relative) improvement from 0.218 to 0.245 macro-F1 score compared to the previous state of the art across 3,912 codes. Finally, we analyse the labelling inconsistencies arising from different coding practices which limit performance on this task.

microRNAs (miRNAs) are major regulators of protein synthesis in the brain. A major goal is to identify changes in miRNA expression underlying protein synthesis-dependent forms of synaptic plasticity such as long-term potentiation (LTP). Previous analyses focused on changes in miRNA levels in total lysate samples. Here, we asked whether changes in total miRNA accurately reflect changes in the amount of miRNA bound to Argonaute protein within the miRNA-induced silencing complex (miRISC). Ago2 immunoprecipitation was used to isolate RISC-associated miRNAs following high-frequency stimulation (HFS)-induced LTP in the dentate gyrus of anesthetized rats. Using locked-nucleic acid-based PCR cards for high-throughput screening and independent validation by quantitative TaqMan RT-PCR, we identified differential regulation of Ago2-associated and total miRNA expression. The ratio of Ago2/total miRNA expression was regulated bidirectionally in a miRNA-specific manner and was largely dependent on N-methyl-D-aspartate receptor (NMDA) activation during LTP induction. The present results identify miRNA association with Ago2 as a potential control point in activity-dependent synaptic plasticity in the adult brain. Finally, novel computational analysis for targets of the Ago2-associated miRNAs identifies 21 pathways that are enriched and differentially targeted by the miRNAs including axon guidance, mTOR, MAPK, Ras, and LTP.

Activity-dependent synaptic plasticity involves rapid regulation of neuronal protein synthesis on a timescale of minutes. miRNA function in synaptic plasticity and memory formation has been elucidated by stable experimental manipulation of miRNA expression and activity using transgenic approaches and viral vectors. However, the impact of rapid miRNA modulation on synaptic efficacy is unknown. Here, we examined the effect of acute (12 min), intrahippocampal infusion of a miR-34a antagonist (antimiR) on medial perforant path-evoked synaptic transmission in the dentate gyrus of adult anesthetised rats. AntimiR-34a infusion acutely depressed medial perforant path-evoked field excitatory post-synaptic potentials (fEPSPs). The fEPSP decrease was detected within 9 min of infusion, lasted for hours, and was associated with knockdown of antimir-34a levels. Antimir-34ainduced synaptic depression was sequence-specific; no changes were elicited by infusion of scrambled or mismatch control. The rapid modulation suggests that a target, or set of targets, is regulated by miR-34a. Western blot analysis of dentate gyrus lysates revealed enhanced expression of Arc, a known miR-34a target, and four novel predicted targets (Ctip2, PKI-1α, TCF4 and Ube2g1). Remarkably, antimiR-34a had no effect when infused during the maintenance phase of long-term potentiation. We conclude that miR-34a regulates basal synaptic efficacy in the adult dentate gyrus in vivo. To our knowledge, these in vivo findings are the first to demonstrate acute (< 9 min) regulation of synaptic efficacy in the adult brain by a miRNA.

Training medical image analysis models requires large amounts of expertly annotated data which is time-consuming and expensive to obtain. Images are often accompanied by free-text radiology reports which are a rich source of information. In this paper, we tackle the automated extraction of structured labels from head CT reports for imaging of suspected stroke patients, using deep learning. Firstly, we propose a set of 31 labels which correspond to radiographic findings (e.g. hyperdensity) and clinical impressions (e.g. haemorrhage) related to neurological abnormalities. Secondly, inspired by previous work, we extend existing state-of-the-art neural network models with a label-dependent attention mechanism. Using this mechanism and simple synthetic data augmentation, we are able to robustly extract many labels with a single model, classified according to the radiologist's reporting (positive, uncertain, negative). This approach can be used in further research to effectively extract many labels from medical text.