In Silico Labeling: Predicting Fluorescent Labels in Unlabeled Images.

Microscopy is a central method in life sciences. Many popular methods, such as antibody labeling, are used to add physical fluorescent labels to specific cellular constituents. However, these approaches have significant drawbacks, including inconsistency; limitations in the number of simultaneous labels because of spectral overlap; and necessary perturbations of the experiment, such as fixing the cells, to generate the measurement. Here, we show that a computational machine-learning approach, which we call "in silico labeling" (ISL), reliably predicts some fluorescent labels from transmitted-light images of unlabeled fixed or live biological samples. ISL predicts a range of labels, such as those for nuclei, cell type (e.g., neural), and cell state (e.g., cell death). Because prediction happens in silico, the method is consistent, is not limited by spectral overlap, and does not disturb the experiment. ISL generates biological measurements that would otherwise be problematic or impossible to acquire.

Three-way comparative study of endoscopic ultrasound-guided transmural gallbladder drainage using lumen-apposing metal stents versus endoscopic transpapillary drainage versus percutaneous cholecystostomy for gallbladder drainage in high-risk surgical patients with acute cholecystitis: clinical outcomes and success in an International, Multicenter Study.

BACKGROUND: Percutaneous cholecystostomy tube (PTGBD), endoscopic retrograde cholangiopancreatography with transpapillary gallbladder drainage (TP), and endoscopic ultrasound-guided transmural gallbladder drainage (EGBD) using lumen-apposing metal stents (LAMS) have been offered for gallbladder decompression for acute cholecystitis in high-risk surgical patients. Yet, there are limited data comparing these therapies. Our aim was to compare the safety and efficacy of EGBD to TP and PTGBD for gallbladder drainage. METHODS: We retrospectively collected high-risk surgical patients from six centers with acute cholecystitis who underwent gallbladder drainage by EGBD, TP, or PTGBD. Data included technical success (gallbladder drainage), clinical success (acute cholecystitis resolution), adverse events (AE), and follow-up. RESULTS: From 2010 to 2016, 372 patients underwent gallbladder drainage, with 146 by PTGBD, 124 by TP, and 102 drained by EGBD. Technical (98% vs. 88% vs. 94%; p = 0.004) and Clinical (97% vs. 90% vs. 80%; p < 0.001) success rates were significantly higher with PTGBD and EGBD compared to TP. PTGBD group had statistically significantly higher number of complications as compared to EGBD and TP groups (2 0% vs. 2% vs. 5%; p = 0.01). Mean hospital stay in the EGBD group was significantly less than TP and PTGBD (16 vs. 18 vs. 19 days; p = 0.01), while additional surgical intervention was significantly higher in the PTGBD group compared to the EGBD and TP groups (49% vs. 4% vs. 11%; p < 0.0001). CONCLUSIONS: EGBD with LAMS is an effective and safer alternative to TP and PTGBD for treatment of patients with acute cholecystitis who cannot undergo surgery. EGBD with LAMS has significantly lower overall AEs, hospital stay, and unplanned admissions compared to PTGBD. TRIAL REGISTRATION: ClinicalTrials.gov Identifier: NCT01522573.

Assessing microscope image focus quality with deep learning.

BACKGROUND: Large image datasets acquired on automated microscopes typically have some fraction of low quality, out-of-focus images, despite the use of hardware autofocus systems. Identification of these images using automated image analysis with high accuracy is important for obtaining a clean, unbiased image dataset. Complicating this task is the fact that image focus quality is only well-defined in foreground regions of images, and as a result, most previous approaches only enable a computation of the relative difference in quality between two or more images, rather than an absolute measure of quality. RESULTS: We present a deep neural network model capable of predicting an absolute measure of image focus on a single image in isolation, without any user-specified parameters. The model operates at the image-patch level, and also outputs a measure of prediction certainty, enabling interpretable predictions. The model was trained on only 384 in-focus Hoechst (nuclei) stain images of U2OS cells, which were synthetically defocused to one of 11 absolute defocus levels during training. The trained model can generalize on previously unseen real Hoechst stain images, identifying the absolute image focus to within one defocus level (approximately 3 pixel blur diameter difference) with 95% accuracy. On a simpler binary in/out-of-focus classification task, the trained model outperforms previous approaches on both Hoechst and Phalloidin (actin) stain images (F-scores of 0.89 and 0.86, respectively over 0.84 and 0.83), despite only having been presented Hoechst stain images during training. Lastly, we observe qualitatively that the model generalizes to two additional stains, Hoechst and Tubulin, of an unseen cell type (Human MCF-7) acquired on a different instrument. CONCLUSIONS: Our deep neural network enables classification of out-of-focus microscope images with both higher accuracy and greater precision than previous approaches via interpretable patch-level focus and certainty predictions. The use of synthetically defocused images precludes the need for a manually annotated training dataset. The model also generalizes to different image and cell types. The framework for model training and image prediction is available as a free software library and the pre-trained model is available for immediate use in Fiji (ImageJ) and CellProfiler.

ALERT-RA: an aperture library-enabled real-time respiratory motion adaptive framework for 4D-VMAT.

PURPOSE: To develop a framework for robust optimization of real-time respiratory motion adaptive VMAT treatment plans, and to evaluate the robustness of resulting plans to variations in tumor trajectory during delivery. METHODS: The proposed framework is called aperture library-enabled real-time robust adaptation (ALERT-RA). A patient-specific library of optimized MLC apertures is defined for each combination of gantry angle and respiratory phase. The method assumes that the tumor is tracked in real-time throughout delivery, and the aperture corresponding to the current phase and gantry angle will be delivered. The aperture library is optimized by considering all possible tumor trajectories determined by a probabilistic respiratory motion model. Plan robustness to trajectory variations was evaluated by sampling a trajectory, and determining the corresponding dose, from the respiratory model for each fraction. The cumulative dose of the full treatment course was simulated 50 times. Percentile dose-volume histograms (PDVHs) were computed from these simulated treatments. The resulting plan quality and robustness of this method were compared to other previously published motion 4D-VMAT methods, including: an optimized tracking approach that assumes reproducible tumor motion, conformal tracking with aperture deformation, and a motion-encompassing method. Two fractionation schemes were tested to determine the possible effect on robustness: a conventional fractionation of 66 Gy in 33 fractions, and an SBRT course with 60 Gy in 5 fractions. RESULTS: When considering target coverage, the ALERT-RA method was found to produce a plan which was more robust than those produced using the optimized or conformal tracking methods. Using the PDVH analysis, the 5th and 95th percentiles of the prescription dose volume for the conventionally fractioned plan were found to be (respectively) 79% and 82% for the optimized tracking approach, 81% and 83% for the conformal tracking approach, and 92% and 97% using the new ALERT-RA method. The motion-encompassing plan was slightly more robust than the ALERT-RA plan, with 5th and 95th percentiles at 94% and 95%, respectively. This came at a cost of higher dose to OARs, with the volume of lung receiving 5 Gy or more equal to 48% for the motion-encompassing plan versus 44% for the ALERT-RA plan. For the SBRT plan, the conformal tracking plan was similarly not robust, with 5th and 95th percentiles of the prescription dose volume equal to 88% and 89%. The optimized tracking SBRT plan gave values of 93% and 95%, and the motion-encompassing plan 94% and 95%, while the ALERT-RA gave values of 93% and 96%. The volume of lung receiving 20 Gy or more was slightly higher for the optimized tracking and motion-encompassing plans compared to the ALERT-RA plan, at 15%, 15%, and 14%, respectively. CONCLUSIONS: Compared to other motion-adaptive VMAT approaches, the ALERT-RA algorithm is capable of delivering high-quality plans which are robust to variations in tumor motion trajectories.

Incidence and clinical significance of epidural fluid signal on spinal magnetic resonance imaging after image-guided lumbar puncture in adults.

PURPOSE: We determined the incidence of epidural fluid signal on spinal magnetic resonance imaging (MRI) after image-guided lumbar puncture (LP) in adults. We correlated those imaging findings with clinical status. METHODS: We searched our institution's medical records from January 2013 through April 2020 to identify adult patients who underwent image-guided LP and postdural puncture MRI. We examined the incidence of epidural fluid signal intensity in adults after image-guided dural puncture, characterized its imaging features, and evaluated its associated clinical factors. RESULTS: Of 91 patients who underwent image-guided dural puncture and subsequent spinal MRI within 7 days, 18 (20%) demonstrated epidural fluid signal on postdural puncture MRI. Univariate analysis showed that larger needle size correlated with signal presence (4/8 [50%] LP with a 20-gauge needle vs 13/82 [16%] with a 22-gauge needle, P = 0.03). The odds of observing epidural fluid signal on postdural puncture MRI decreased by 8% per 1-year increase in age (P < 0.001). Postdural puncture symptoms did not differ between those with and without epidural fluid signal intensity. CONCLUSION: Epidural fluid signal on MRI in adults may be an incidental finding more commonly observed in younger patients after dural puncture with larger needles, without a significant correlation with symptomatology.

The anterior head of deltoid in relation to anterior clavicle plating: how much are we releasing?

Middle to distal-third clavicular shaft fractures are commonly treated with precontoured anterior plating. Some surgeons use mini-fragment plate fixation and position these plates on the anterior clavicle. Recent studies demonstrated the advantages of anterior clavicle plating, including a possible biomechanical advantage with cantilever bending forces and less subsequent implant removal. The insertion and positioning of anteriorly based clavicle plates requires the release of a portion of the anterior deltoid origin from the lateral clavicle. The purpose of this study is to evaluate the anatomy of the deltoid in relation to the clavicle and to determine the percentage of the deltoid origin released to place modern anterior precontoured plates. Methods: Six right and 4 left cadaver shoulders were dissected, each from separate cadaveric specimens (6 male and 4 female). All measurements were made with digital calipers. The length of the clavicle was measured from the acromioclavicular joint to the sternoclavicular joint. The length of deltoid origin on the lateral clavicle was measured from the acromioclavicular joint to the most medial attachment of the deltoid on the clavicle. Percentage of clavicle with deltoid origin was subsequently calculated. Results: The average length of the cadaveric clavicles was 164.4 mm with a range from 134.3 to 178.1 mm. The average amount of deltoid origin on the clavicle was 58.7 mm with a range from 43.4 to 69 mm. On average 35.5% of the clavicle had deltoid origin, with a range from 30.2% to 38.8%. Conclusion: On average, 35.5% of the clavicular osseous anatomy contains deltoid origin. This should be taken into consideration when performing anterior plating for clavicle fractures. With a significant portion of deltoid origin elevated, surgeons may consider altering postoperative protocols until some interval healing has occurred to this anterior head of the deltoid.

The prevalence of undiagnosed diabetes in non-cardiac surgery patients, an observational study.

PURPOSE: Given that preoperative hyperglycemia is associated with poor outcomes and many non-diabetic patients have high plasma glucose (PG) levels, the purpose of our study was to estimate the prevalence of undiagnosed diabetes among non-cardiac surgery patients and to identify predictors of hyperglycemia in non-diabetics. METHODS: We included all non-cardiac surgery patients with complete records in the Clinical Database of the Anesthesiology Institute at the Cleveland Clinic during January 2007 to April 2009, and we estimated the prevalence of undiagnosed diabetes and impaired fasting glucose (IFG) among the non-diabetic patients. The mean glucose levels for known diabetics and undiagnosed diabetics were compared using two-tailed Student's t tests, and we assessed the association between PG levels and demographic variables within the non-diabetics. RESULTS: Of the 39,434 patients analyzed, 5,511 (14%) were known diabetics. Of the 33,923 known non-diabetics, 3,426 (10 %) were undiagnosed diabetics and another 3,549 (11%) had IFG. Thus, 6,975 patients (21%) of the non-diabetic patients presented with abnormally high glucose. Previously undiagnosed diabetics had higher preoperative glucose levels compared with known diabetics, with a mean ± standard deviation (SD) of 161 ± 48 vs 146 ± 67 mg·dL⁻¹ (8.9 ± 2.7 vs 8.1 ± 3.7 mmoL·L⁻¹), respectively. The difference remained highly significant after adjusting for body mass index, age, sex, and American Society of Anesthesiologists (ASA) physical status (P < 0.001). Among non-diabetics, older age, obesity, male sex, and a higher ASA physical status were collectively significant predictors of hyperglycemia, with a c-statistic (95% confidence interval) of 0.67 (0.66-0.68). CONCLUSION: A significant proportion of non-cardiac surgery patients have previously undiagnosed diabetes and pre-diabetes. Previously undiagnosed patients have higher fasting glucose levels compared with diabetic patients. Further studies should be conducted to identify the implications of these findings on patient outcomes.

Pneumatosis Intestinalis in Adult Bilateral Lung Transplant Patients: A Single Institution Experience and Literature Review.

Pneumatosis intestinalis (PI) is a radiologic finding which is characterized by the accumulation of gas within the bowel wall. This radiologic finding is traditionally thought of in the sense of intestinal ischemia. An uncommon cause of this finding is post organ transplantation. We did an institutional and literature review of this finding to demonstrate its distinct imaging features and benign nature. It was observed to occur in approximately 5.2% of patients post lung transplant (23/442). On imaging, it displays an expansile/bubbly appearance of gas within the bowel wall that is distinct from the traditional findings seen in intestinal ischemia. Clinical review showed that posttransplant patients with PI can be successfully managed conservatively with early enteral nutrition, oxygen, antibiotics, and limited follow-up imaging. With the increasing use of organ transplantation, PI is being diagnosed with increased frequency. It is important to let clinicians know of this entity and its potential outcomes.

The roles of visual expertise and visual input in the face inversion effect: behavioral and neurocomputational evidence.

Research has shown that inverting faces significantly disrupts the processing of configural information, leading to a face inversion effect. We recently used a contextual priming technique to show that the presence or absence of the face inversion effect can be determined via the top-down activation of face versus non-face processing systems [Ge, L., Wang, Z., McCleery, J., & Lee, K. (2006). Activation of face expertise and the inversion effect. Psychological Science, 17(1), 12-16]. In the current study, we replicate these findings using the same technique but under different conditions. We then extend these findings through the application of a neural network model of face and Chinese character expertise systems. Results provide support for the hypothesis that a specialized face expertise system develops through extensive training of the visual system with upright faces, and that top-down mechanisms are capable of influencing when this face expertise system is engaged.

Continuous aperture dose calculation and optimization for volumetric modulated arc therapy.

Although VMAT delivery features continuous gantry rotation and leaf motion, dose calculation is often performed under the dual assumption of discrete apertures changing instantaneously from one discrete angle to the next. In this work, the validity of these two approximations is determined, as well as their impact on the quality of optimized plans. Further, an accurate method of fluence calculation is derived which does not use the discrete aperture approximation, but instead calculates the fluence as the multi-leaf collimator leaves sweep from one position to another. This continuous aperture fluence calculation is integrated in the VMAT optimization process using the open-source treatment planning system matRad. The three-step approach of VMAT optimization is used: fluence map optimization followed by leaf sequencing and direct aperture optimization, with variable leaf speed, gantry rotation speed, and MU rate. The benefit of the continuous aperture VMAT method over the discrete aperture method is determined by comparing the plan quality of discrete aperture and continuous aperture optimized plans, when the former is recalculated using the continuous aperture fluence calculation. Discrete aperture VMAT plans calculated at 4° spacing result in significant dose errors (10%-35%, depending on the anatomical site) as compared to the reference dose (continuous aperture fluence calculation at 0.5° spacing). These errors are greatly reduced (to 0.8%-2%) when the continuous aperture fluence calculation method was used at the same 4° spacing, implying that the dose error is primarily due to the discrete aperture approximation. Whereas all dose objectives were met by the discrete aperture VMAT optimized plan, many of them failed when the dose was recalculated with the continuous aperture fluence calculation. All objectives were met once again when the plan was optimized with the new continuous aperture VMAT optimization. Further, using only half of the beam angles, the continuous aperture VMAT optimization can achieve the same degree of accuracy with only 40% of the computing time as compared with the standard discrete aperture VMAT.

Learning grammatical structure with Echo State Networks.

Echo State Networks (ESNs) have been shown to be effective for a number of tasks, including motor control, dynamic time series prediction, and memorizing musical sequences. However, their performance on natural language tasks has been largely unexplored until now. Simple Recurrent Networks (SRNs) have a long history in language modeling and show a striking similarity in architecture to ESNs. A comparison of SRNs and ESNs on a natural language task is therefore a natural choice for experimentation. Elman applies SRNs to a standard task in statistical NLP: predicting the next word in a corpus, given the previous words. Using a simple context-free grammar and an SRN with backpropagation through time (BPTT), Elman showed that the network was able to learn internal representations that were sensitive to linguistic processes that were useful for the prediction task. Here, using ESNs, we show that training such internal representations is unnecessary to achieve levels of performance comparable to SRNs. We also compare the processing capabilities of ESNs to bigrams and trigrams. Due to some unexpected regularities of Elman's grammar, these statistical techniques are capable of maintaining dependencies over greater distances than might be initially expected. However, we show that the memory of ESNs in this word-prediction task, although noisy, extends significantly beyond that of bigrams and trigrams, enabling ESNs to make good predictions of verb agreement at distances over which these methods operate at chance. Overall, our results indicate a surprising ability of ESNs to learn a grammar, suggesting that they form useful internal representations without learning them.