Editorial: Hierarchical Object Representations in the Visual Cortex and Computer Vision

Improving CT Image Tumor Segmentation Through Deep Supervision and Attentional Gates

Computer Tomography (CT) is an imaging procedure that combines many X-ray measurements taken from different angles. The segmentation of areas in the CT images provides a valuable aid to physicians and radiologists in order to better provide a patient diagnose. The CT scans of a body torso usually include different neighboring internal body organs. Deep learning has become the state-of-the-art in medical image segmentation. For such techniques, in order to perform a successful segmentation, it is of great importance that the network learns to focus on the organ of interest and surrounding structures and also that the network can detect target regions of different sizes. In this paper, we propose the extension of a popular deep learning methodology, Convolutional Neural Networks (CNN), by including deep supervision and attention gates. Our experimental evaluation shows that the inclusion of attention and deep supervision results in consistent improvement of the tumor prediction accuracy across the different datasets and training sizes while adding minimal computational overhead.

Dynamic distractor environments reveal classic visual field anisotropies for judgments of temporal order

Numerous studies have shown that visual performance critically depends on the stimulus' projected retinal location. For example, performance tends to be better along the horizontal relative to the vertical meridian (lateral anisotropy). Another case is the so-called upper-lower anisotropy, whereby performance is better in the upper relative to the lower hemifield. This study investigates whether temporal order judgments (TOJs) are subject to these visual field constraints. In Experiments 1 and 2, subjects reported the temporal order of two disks located along the horizontal or vertical meridians. Each target disk was surrounded by 10 black and white distractor disks, whose polarity remained unchanged (static condition) or reversed throughout the trial (dynamic condition). Results indicate that the mere presence of dynamic distractors elevated thresholds by more than a factor of four and that this elevation was particularly pronounced along the vertical meridian, evidencing the lateral anisotropy. In Experiment 3, thresholds were compared in upper, lower, left, and right visual hemifields. Results show that the threshold elevation caused by dynamic distractors was greatest in the upper visual field, demonstrating an upper-lower anisotropy. Critically, these anisotropies were evident exclusively in dynamic distractor conditions suggesting that distinct processes govern TOJ performance under these different contextual conditions. We propose that whereas standard TOJs are processed by fast low-order motion mechanisms, the presence of dynamic distractors mask these low-order motion signals, forcing observers to rely more heavily on more sluggish higher order motion processes.