Deep learning techniques have enabled rapid progress in monocular depth estimation, but their quality is limited by the ill-posed nature of the problem and the scarcity of high quality datasets. We estimate depth from a single camera by leveraging the dual-pixel auto-focus hardware that is increasingly common on modern camera sensors. Classic stereo algorithms and prior learning-based depth estimation techniques underperform when applied on this dualpixel data, the former due to too-strong assumptions about RGB image matching, and the latter due to not leveraging the understanding of optics of dual-pixel image formation. To allow learning based methods to work well on dual-pixel imagery, we identify an inherent ambiguity in the depth estimated from dual-pixel cues, and develop an approach to estimate depth up to this ambiguity. Using our approach, existing monocular depth estimation techniques can be effectively applied to dual-pixel data, and much smaller models can be constructed that still infer high quality depth. To demonstrate this, we capture a large dataset of in-the-wild 5-viewpoint RGB images paired with corresponding dualpixel data, and show how view supervision with this data can be used to learn depth up to the unknown ambiguity. On our new task, our model is 30% more accurate than any prior work on learning-based monocular or stereoscopic depth estimation.

No abstract

No abstract

We present a method for precisely time-synchronizing the capture of image sequences from a collection of smartphone cameras connected over WiFi. Our method is entirely software-based, has only modest hardware requirements, and achieves an accuracy of less than 250 µs on unmodified commodity hardware. It does not use image content and synchronizes cameras prior to capture. The algorithm operates in two stages. In the first stage, we designate one device as the leader and synchronize each client device's clock to it by estimating network delay. Once clocks are synchronized, the second stage initiates continuous image streaming, estimates the relative phase of image timestamps between each client and the leader, and shifts the streams into alignment. We quantitatively validate our results on a multi-camera rig imaging a high-precision LED array and qualitatively demonstrate significant improvements to multi-view stereo depth estimation and stitching of dynamic scenes. We release as open source libsoftwaresync, an Android implementation of our system, to inspire new types of collective capture applications.

Isolated third cranial nerve palsies may be caused by compressive intracranial aneurysms located at the junction of the internal carotid and posterior communicating arteries or, less commonly, at the apex of the basilar artery or its junction with the superior cerebellar or posterior cerebral arteries. Such aneurysms typically measure at least 4 mm in diameter. Technical improvements in noninvasive techniques, including CT and MRA, have yielded a detection rate of such aneurysms that approaches that of catheter cerebral angiography (CCA), which itself carries a small but serious risk. Multidetector technology, which allows a rapid scan time, has promoted CT to the first choice for investigating aneurysms in this setting except when dye or radiation exposure is unacceptable, as with pregnant women, children, and those with renal or severe cardiac disease. Major impediments to accurate detection are a lack of availability of trained technicians, who must perform manipulation of the raw imaging data ("post-processing"), and a paucity of certified neuroradiologists with the time, skill, and experience to devote to interpreting difficult cases. To avoid diagnostic mishaps, noninvasive studies should be reviewed by at least one neuroradiologist before aneurysm is rejected as the cause or before the patient undergoes CCA.

Background: Contrast-induced neurotoxicity (CIN) is a rare complication of neurointerventional procedures and its understanding remains limited. We evaluated the association of CIN with systemic hemodynamics in patients undergoing neuroendovascular interventions. Methods: We conducted a 1:2 matched case-control study from a prospectively collected database of 2510 neurointerventional patients. We defined CIN as new neurological deficits presented ≤24 h post-operation after excluding other possible etiologies. We obtained demographic, clinical and imaging data, and baseline and intraprocedural blood pressures (BP) from medical records. The area between baseline and intraprocedural BP was used to measure sustained variability of BP over time. A generalized linear mixed model and generalized estimating equation were used to analyze the BP difference between groups over time. Results: We evaluated 11 CIN cases and 22 controls. 2746 and 5837 min of continued BP data were analyzed for cases and controls, respectively. CIN cases had higher measurements and greater variability for: Systolic BP (SBP) [median 125 (IQR:121-147) vs. 114 (IQR:107-124) mmHg], median area above baseline [median 350 (IQR:25-1328) vs. 52 (IQR:0-293) mmHg*minutes] and mean arterial pressure (MAP) [median 85 (IQR:79-98) vs. 80 (IQR:74-89) mmHg]. CIN cases demonstrated a significant mean increase in SBP and MAP of 23.41 mmHg (p < 0.01) and 13.79 mmHg (p < 0.01) when compared to controls, respectively, over the perioperative time. Conclusion: Sustained hypertension and high BP variability may contribute to the pathophysiology of CIN. Acute hypertension can increase blood-brain barrier permeability and potentially allow contrast to leak into the brain parenchyma causing direct toxicity and CIN symptoms.

Objectives: The objective was to evaluate safety of intravenous (IV) tissue plasminogen activator (tPA) delivered without dedicated thrombolytic stroke teams.Methods: This was a retrospective, observational study of patients treated between 1996 and 2005 at four southeastern Michigan hospital emergency departments (EDs) with a prospectively defined comparison to the National Institute of Neurological Disorders and Stroke (NINDS) tPA stroke study cohort. Main outcome measures were mortality, intracerebral hemorrhage (ICH), systemic hemorrhage, neurologic recovery, and guideline violations.Results: A total of 273 consecutive stroke patients were treated by 95 emergency physicians (EPs) using guidelines and local neurology resources. One-year mortality was 27.8%. Unadjusted Cox model relative risk (RR) of mortality compared to the NINDS tPA treatment and placebo groups was 1.20 (95% confidence interval [CI] = 0.87 to 1.64) and 1.04 (95% CI = 0.76 to 1.41), respectively. The rate of significant ICH by computed tomography (CT) criteria was 6.6% (odds ratio [OR] = 1.03, 95% CI = 0.56 to 1.90 compared to the NINDS tPA treatment group). The proportions of symptomatic ICH by two other prespecified sets of clinical criteria were 4.8 and 7.0%. The rate of any ICH within 36 hours of treatment was 9.9% (RR = 0.94, 95% CI = 0.58 to 1.51 compared to the NINDS tPA group). The occurrence of major systemic hemorrhage (requiring transfusion) was 1.1%. Functional recovery by the modified Rankin Scale score (mRS = 0 to 2) at discharge occurred in 38% of patients with a premorbid disability mRS < 2. Guideline deviations occurred in the ED in 26% of patients and in 25% of patients following admission.Conclusions: In these EDs there was no evidence of increased risk with respect to mortality, ICH, systemic hemorrhage, or worsened functional outcome when tPA was administered without dedicated thrombolytic stroke teams. Additional effort is needed to improve guideline compliance.

No abstract