Automated Detection of Candidate Subjects with Cerebral Microbleeds using Machine Learning

Cerebral microbleeds (CMBs) appear as small, circular, well defined hypointense lesions of a few mm in size on T2*-weighted gradient recalled echo (T2*-GRE) images and appear enhanced on susceptibility weighted images (SWI). Due to their small size, contrast variations and other mimics (e.g. blood vessels), CMBs are highly challenging to detect automatically. In large datasets (e.g. the UK Biobank dataset), exhaustively labelling CMBs manually is difficult and time consuming. Hence it would be useful to preselect candidate CMB subjects in order to focus on those for manual labelling, which is essential for training and testing automated CMB detection tools on these datasets. In this work, we aim to detect CMB candidate subjects from a larger dataset, UK Biobank, using a machine learning-based, computationally light pipeline. For our evaluation, we used 3 different datasets, with different intensity characteristics, acquired with different scanners. They include the UK Biobank dataset and two clinical datasets with different pathological conditions. We developed and evaluated our pipelines on different types of images, consisting of SWI or GRE images. We also used the UK Biobank dataset to compare our approach with alternative CMB preselection methods using non-imaging factors and/or imaging data. Finally, we evaluated the pipeline’s generalisability across datasets. Our method provided subject-level detection accuracy > 80% on all the datasets (withindataset results), and showed good generalisability across datasets, providing a consistent accuracy of over 80%, even when evaluated across different modalities.

Evaluation and Comparison of Artificial Intelligence Vision Aids: Orcam MyEye 1 and Seeing AI

Introduction: We compared the print-to-speech properties and human performance characteristics of two artificial intelligence vision aids, Orcam MyEye 1 (a portable device) and Seeing AI (an iPhone and iPad application). Methods: There were seven participants with visual impairments who had no experience with the two reading aids. Four participants had no light perception. Two individuals with measurable acuity and one with light perception were tested while blindfolded. We also tested performance with text of varying appearance in varying viewing conditions. To evaluate human performance, we asked the participants to use the devices to attempt 12 reading tasks similar to activities of daily living. We assessed the ranges of text attributes for which reading was possible, such as print size, contrast, and light level. We also assessed if individuals could complete tasks with the devices and measured accuracy and completion time. Participants also completed a survey concerning the two aids. Results: Both aids achieved greater than 95% accuracy in text recognition for flat, plain word documents and ranged from 13 to 57% accuracy for formatted text on curved surfaces. Both aids could read print sizes as small as 0.8M (20/40 Snellen equivalent, 40 cm viewing distance). Individuals successfully completed 71% and 55% ( p = .114) of tasks while using Orcam MyEye 1 and Seeing AI, respectively. There was no significant difference in time to completion of tasks ( p = .775). Individuals believed both aids would be helpful for daily activities. Discussion: Orcam MyEye 1 and Seeing AI had similar text-reading capability and usability. Both aids were useful to users with severe visual impairments in performing reading tasks. Implications for Practitioners: Selection of a reading device or aid should be based on individual preferences and prior familiarity with the platform, since we found no clear superiority of one solution over the other.

COVD-29. CONTINUATION OF TEMOZOLOMIDE CHEMOTHERAPY IN A GLIOBLASTOMA PATIENT AFTER RESOLUTION OF COVID-19 PNEUMONIA

We present a case of a 33 year-old patient with glioblastoma (IDH wild type, MGMT unmethylated) who was diagnosed with COVID-19 pneumonia while undergoing chemotherapy. The patient did not have any medical comorbidities. He was clinically asymptomatic following surgery, completed concurrent phase of combined chemotherapy and radiation and was undergoing treatment with adjuvant temozolomide. He had radiographic improvement of the brain tumor (decreased size, contrast enhancement and T2 flair) after three cycles of adjuvant temozolomide. However, after cycle three the patient developed fever and abdominal pain. Evaluation in the emergency room revealed low absolute lymphocyte count (0.7 K/MM3), positive COVID-19 point of care test and CT chest revealed patchy peripheral bibasilar ground glass and consolidative opacities compatible with pulmonary infection, with viral etiology such as COVID. Symptoms resolved after 2 weeks. Due to active infection and leucopenia temozolomide was on hold for 1 month. He was considered cleared of infection after resolution of symptoms. Temozolomide was initiated after resolution of leucopenia. Patient continued to do well after administration of subsequent temozolomide cycles and repeat CT chest after 2 months revealed resolution of consolidation and no new areas of consolidation. Temozolomide was safely administered in this patient without reactivation of COVID-19 infection. He did not have any thrombotic events.

Combining of objects images under conditions of inaccurate description of their images

An attribute of modern robotic systems and control systems for unmanned vehicles is a machine vision, which is based on processing of images obtained from highly informative radar or optoelectronic sensors. Effective image processing and recognition procedures are based on use of morphological features of objects images with high space‑time stability. This paper considers one of the aspects of images morphological analysis related to the influence of errors in the reference contour descriptions of objects. With use of developed analytical apparatus it is shown that the result of such errors is a decreasing of object/background contrast and, as a result, increasing of image comparison errors. The degree of this increase depends on the object type (area, linear) and its characteristics (size, contrast). Practical recommendations on the acceptable level of errors in the representation of object descriptions are justified.

Ornamental and landscape potential of a bromeliad native to the Cerrado

Bromelia reversacantha Mez is an endemic species of the Cerrado Biome, for which there is no published data about its ornamental potentiality. The objective was to identify the ornamental and landscape potential of this bromeliad native from Cerrado. In September 2015, researchers specialized in the field of native plants evaluated the vegetative, reproductive and aesthetic characteristics of twenty plants of natural occurrence in the Biological Reserve Prof. José Ângelo Rizzo, in Mossâmedes, GO. In a scale of zero to ten, the bromeliad received an average score of 7.1 for the ornamental aspect, being 7.4; 7.3 and 6.6 respectively for the flowering plant, plant with fruits, and plant in the vegetative stage. The vigorous size, contrast, and color of the leaves and bracts, the showy inflorescence, the flowers, and fruits were shown as ornamental attributes. Bromelia reversacantha presents several characteristics favorable to its introduction in landscape projects, suggesting the use in gardens, or as living fences/borders and in pots; as well as in ornamentation in decorative arrangements with the use of inflorescences and fruit bunches.

The Effect of Visual Function on the Batting Performance of Professional Baseball Players

This report evaluates the role of the combined visual abilities of acuity, contrast sensitivity and presentation time on plate discipline and baseball batting performance. A visual function test (EVTS) was performed on 585 professional baseball players. The results were compared to several common plate-discipline measures. The EVTS test provides a single measure combining target size, contrast and presentation time. Correlations (statistically significant) were found between this measure and several plate discipline metrics (InzoneSwingPct, inzoneFbSwingPct, ChasePct, FbChasePct, BBperPa). Years of major league service did not appear to be related to visual ability. When comparing the best and worst 20% groups based on visual ability, statistically significant improvements ranging from 11.6% in BBperPa to 3.5% in inzoneSwingPct were noted in the better visual function group. Effect sizes ranged from 0.278 to 0.387. These results demonstrate the relationship between basic visual function and batting performance. These are the first results, on a large group at the professional level, to demonstrate this relationship statistically. These results may aid player selection, indicating that batters with better visual function are more likely to be successful when batting and more productive for their team.

Distinct patterns of surround modulation in V1 and hMT+

Modulation of a neuron’s responses by the stimuli presented outside of its classical receptive field is ubiquitous in the visual system. This “surround modulation” mechanism is believed to be critical for efficient processing and leads to many well-known perceptual effects. The details of surround modulation, how-ever, are still not fully understood. One of the open questions is related to the differences in surround modulation mechanisms in different cortical areas, and their interactions. Here we study patterns of surround modulation in primary visual cortex (V1) and middle temporal complex (hMT+) utilizing a well-studied effect in motion perception, where human observers’ ability to discriminate the drift direction of a grating improves as its size gets bigger if the grating has a low contrast, and deteriorates if it has a high contrast. We first replicated the findings in the literature with a behavioral experiment using small and large (1.06 and 8.05 degrees of visual angle) drifting gratings with either low (2%) or high (99%) contrast presented at the periphery. Next, using functional MRI, we found that in V1 with increasing size cortical responses increased at both contrast levels, but they increased more at high contrast. Whereas in hMT+ with increasing size cortical responses remained unchanged at high contrast, and increased at low contrast. These findings show that surround modulation in V1 and hMT+ are distinct. Furthermore these findings provide evidence that the size-contrast interaction in motion perception is likely to originate in hMT+.

Cross-modal size-contrast illusion: Acoustic increases in intensity and bandwidth modulate haptic representation of object size

Changes in the retinal size of stationary objects provide a cue to the observer’s motion in the environment: Increases indicate the observer’s forward motion, and decreases backward motion. In this study, a series of images each comprising a pair of pine-tree figures were translated into auditory modality using sensory substitution software. Resulting auditory stimuli were presented in an ascending sequence (i.e. increasing in intensity and bandwidth compatible with forward motion), descending sequence (i.e. decreasing in intensity and bandwidth compatible with backward motion), or in a scrambled order. During the presentation of stimuli, blindfolded participants estimated the lengths of wooden sticks by haptics. Results showed that those exposed to the stimuli compatible with forward motion underestimated the lengths of the sticks. This consistent underestimation may share some aspects with visual size-contrast effects such as the Ebbinghaus illusion. In contrast, participants in the other two conditions did not show such magnitude of error in size estimation; which is consistent with the “adaptive perceptual bias” towards acoustic increases in intensity and bandwidth. In sum, we report a novel cross-modal size-contrast illusion, which reveals that auditory motion cues compatible with listeners’ forward motion modulate haptic representations of object size.

Hazard Detection Prediction Model for Rural Roads Based on Hazard and Environment Properties

A driver’s reaction time encountering hazards on roads involves different sections, and each section must occur at the right time to prevent a crash. An appropriate reaction starts with hazard detection. A hazard can be detected on time if it is completely visible to the driver. It is assumed in this paper that hazard properties such as size and color, the contrast between the environment and a hazard, whether the hazard is moving or fixed, and the presence of a warning are effective in improving driver hazard detection. A driving simulator and different scenarios on a two-lane rural road are used for assessing novice and experienced drivers’ hazard detection, and a Sugeno fuzzy model is used to analyze the data. The results show that the hazard detection ability of novice and experienced drivers decreases by 35% and 64%, respectively, during nighttime compared to daytime. Also, moving hazards increase hazard detection ability by 9% and 180% for experienced and novice drivers, respectively, compared to fixed hazards. Moreover, increasing size, contrast, and color difference affect hazard detection under nonlinear functions. The results could be helpful in safety improvement solution prioritization and in preventing vehicle-pedestrian, vehicle-animal, and vehicle-object crashes, especially for novice drivers.