Search

The use of deep neural networks (DNNs) in safety-critical applications like mobile health and autonomous driving is challenging due to numerous model-inherent shortcomings. These shortcomings are diverse and range from a lack of generalization over insufficient interpretability to problems with malicious inputs. Cyber-physical systems employing DNNs are therefore likely to suffer from safety concerns. In recent years, a zoo of state-of-the-art techniques aiming to address these safety concerns has emerged. This work provides a structured and broad overview of them. We first identify categories of insufficiencies to then describe research activities aiming at their detection, quantification, or mitigation. Our paper addresses both machine learning experts and safety engineers: The former ones might profit from the broad range of machine learning topics covered and discussions on limitations of recent methods. The latter ones might gain insights into the specifics of modern ML methods. We moreover hope that our contribution fuels discussions on desiderata for ML systems and strategies on how to propel existing approaches accordingly., 94 pages

Deep neural networks (DNNs) for highly automated driving are often trained on a large and diverse dataset, and evaluation metrics are reported usually on a per-frame basis. However, when evaluated on video sequences, the predictions are often unstable between consecutive frames. As such unstable predictions over time can lead to severe safety consequences, there is a growing need to understand, evaluate, and improve the temporal consistency of DNNs. In this paper, we explore such a temporal characteristic and propose a novel unsupervised temporal consistency (TC) loss that penalizes unstable semantic segmentation predictions. This loss function is used in a two-stage training scheme to jointly optimize for both, accuracy of semantic segmentation predictions, and its temporal consistency based on video sequences. We demonstrate that our training strategy helps in improving the temporal consistency of two state-of-the-art semantic segmentation networks on two different road-scenes datasets. We report an absolute 4.25% improvement in the mean temporal consistency (mTC) of the HRNetV2 network and an absolute 2.78% improvement on the DeepLabv3+ network, both evaluated on the Cityscapes dataset, with only a slight decrease in accuracy. When evaluating on the same video sequences using a synthetic dataset Sim KI-A, we show absolute improvements in both, accuracy (2.19% mIoU) and temporal consistency (0.21% mTC) for the DeepLabv3+ network. We confirm similar improvements for the HRNetV2 network.

Enabling autonomous driving (AD) can be considered one of the biggest challenges in today's technology. AD is a complex task accomplished by several functionalities, with environment perception being one of its core functions. Environment perception is usually performed by combining the semantic information captured by several sensors, i.e., lidar or camera. The semantic information from the respective sensor can be extracted by using convolutional neural networks (CNNs) for dense prediction. In the past, CNNs constantly showed state-of-the-art performance on several vision-related tasks, such as semantic segmentation of traffic scenes using nothing but the red-green-blue (RGB) images provided by a camera. Although CNNs obtain state-of-the-art performance on clean images, almost imperceptible changes to the input, referred to as adversarial perturbations, may lead to fatal deception. The goal of this article is to illuminate the vulnerability aspects of CNNs used for semantic segmentation with respect to adversarial attacks, and share insights into some of the existing known adversarial defense strategies. We aim to clarify the advantages and disadvantages associated with applying CNNs for environment perception in AD to serve as a motivation for future research in this field., Comment: IEEE Signal Processing Magazine (Volume: 38, Issue: 1, Jan. 2021), pp. 42 - 52

Deep neural networks are often not robust to semantically-irrelevant changes in the input. In this work we address the issue of robustness of state-of-the-art deep convolutional neural networks (CNNs) against commonly occurring distortions in the input such as photometric changes, or the addition of blur and noise. These changes in the input are often accounted for during training in the form of data augmentation. We have two major contributions: First, we propose a new regularization loss called feature-map augmentation (FMA) loss which can be used during finetuning to make a model robust to several distortions in the input. Second, we propose a new combined augmentations (CA) finetuning strategy, that results in a single model that is robust to several augmentation types at the same time in a data-efficient manner. We use the CA strategy to improve an existing state-of-the-art method called stability training (ST). Using CA, on an image classification task with distorted images, we achieve an accuracy improvement of on average 8.94% with FMA and 8.86% with ST absolute on CIFAR-10 and 8.04% with FMA and 8.27% with ST absolute on ImageNet, compared to 1.98% and 2.12%, respectively, with the well known data augmentation method, while keeping the clean baseline performance., Accepted at ACM CSCS 2020 (8 pages, 4 figures)

Commonly used metrics to evaluate semantic segmentation such as mean intersection over union (mIoU) do not incorporate temporal consistency. A straightforward extension of existing metrics towards evaluating the consistency of segmentation of video sequences does not exist, since labelled videos are rare and very expensive to obtain. For safety-critical applications such as highly automated driving, there is, however, a need for a metric that measures such temporal consistency of video segmentation networks to possibly support safety requirements. In this paper, (a) we introduce a metric which does not require segmentation labels for measuring the stability of the predictions of segmentation networks over a series of images; (b) we perform an in-depth analysis of the proposed metric and observe strong correlations to the supervised mIoU metric; (c) we perform an evaluation of five state-of-the-art networks for semantic segmentation of varying complexities and architectures evaluated on two public datasets, namely, Cityscapes and CamVid. Finally, we perform timing evaluations and propose the use of the metric as either an online observer for identification of possibly unstable segmentation predictions, or as an offline method to evaluate or to improve semantic segmentation networks, e.g., by selecting additional training data with critical temporal consistency.

Despite recent advancements, deep neural networks are not robust against adversarial perturbations. Many of the proposed adversarial defense approaches use computationally expensive training mechanisms that do not scale to complex real-world tasks such as semantic segmentation, and offer only marginal improvements. In addition, fundamental questions on the nature of adversarial perturbations and their relation to the network architecture are largely understudied. In this work, we study the adversarial problem from a frequency domain perspective. More specifically, we analyze discrete Fourier transform (DFT) spectra of several adversarial images and report two major findings: First, there exists a strong connection between a model architecture and the nature of adversarial perturbations that can be observed and addressed in the frequency domain. Second, the observed frequency patterns are largely image- and attack-type independent, which is important for the practical impact of any defense making use of such patterns. Motivated by these findings, we additionally propose an adversarial defense method based on the well-known Wiener filters that captures and suppresses adversarial frequencies in a data-driven manner. Our proposed method not only generalizes across unseen attacks but also beats five existing state-of-the-art methods across two models in a variety of attack settings., Comment: Accepted by The International Joint Conference on Neural Network (IJCNN) 2021

The prevalence of skeletal dysplasias is between 1 and 2000, and 1 and 4000 live births. While here are over 200 skeletal dysplasias approximately four disorders comprise 70% of the total: Achondroplasia, thanatophoric dysplasia, osteogenesis imperfecta, and achondrogenesis. The appropriate identification of lethal skeletal dysplasia is important not only for current pregnancy management, but also for genetic counseling concerning future pregnancies. Detection of skeletal dysplasias is usually possible by prenatal ultrasound, an accurate specific diagnosis is possible by radiologic, pathologic and molecular genetic examination. A total body ultrasound approach should include assessment of the following: Limbs, long bones and extremities, bone mineralization, any joint contractures, joint dislocations, fetal calvarium, spine and thorax.

Background This retrospective study of prospectively collected data compared coronary artery bypass graft (CABG) surgery to drug‐eluting stenting (DES) in diabetic patients with multivessel coronary artery disease (CAD). Prior randomized trials and clinical studies have suggested that CABG may be the preferred revascularization strategy in diabetic patients with multivessel CAD. Data are limited regarding the impact of DES vs. CABG on clinical outcomes. Methods We included 205 consecutive diabetic patients who underwent either CABG ( n =103) or DES ( n =102). The primary clinical end points were freedom from major adverse cardiac events (MACE) at 30 days and 1 year. Results Baseline characteristics were similar between both groups. At 1 year, the mortality rate was similar in the CABG and DES group (8% vs. 10%, p =0.6) but the MACE rate was lower in the CABG group (12% vs. 27%, p =0.006) due to less repeat revascularization with CABG (3% vs. 20%, p p =0.04). Angiographically‐documented stent thrombosis after DES occurred in 3%. Presentation with acute myocardial infarction (hazard ratio [HR], 2.26, 95% CI, 1.13 to 4.55) and DES (HR, 2.4, 95% CI, 1.23 to 4.77) were positive independent predictors, whereas therapy with a statin was a negative independent predictor of MACE (HR, 0.40, 95% CI, 0.21 to 0.76). Conclusions Bypass surgery was associated with less MACE primarily due to the higher repeat revascularization rate with DES and is therefore superior to DES despite more extensive CAD in CABG patients.

Background Severe aortic stenosis (AS) is a surgically correctable condition. However, aortic valve replacement (AVR) is not offered to many patients with severe AS for various reasons. We investigated the profile and survival patterns of patients with severe AS who did not have AVR. Methods Our echocardiographic database was screened for patients with severe AS, defined as a Doppler estimated aortic valve area of 0.8 cm 2 or less between 1993 and 2003. Seven hundred and forty patients with severe AS were identified, of whom 453 patients had no AVR through the follow-up period, forming the study cohort. These patients were comprehensively characterized by obtaining clinical, pharmacologic, and surgical data through a comprehensive chart review and extracting survival data from the National Death Index. Results Patient characteristics were as follows: age 75 ± 13 years, 48% male, left ventricular (LV) ejection fraction 52 ± 21%, coronary artery disease in 34%, hypertension in 35%, serum creatinine level greater than 2 mg/dL in 11%, and diabetes mellitus in 14%. The survival at 1 year, 5 years, and 10 years was 62%, 32%, and 18%, respectively. The univariate predictors of reduced survival were advanced age, low LV ejection fraction, heart failure, elevated serum creatinine level, severe mitral regurgitation, and pulmonary hypertension; and the independent predictors were advanced age, low LV ejection fraction, heart failure, elevated serum creatinine level, and systemic hypertension. Concomitant pharmacotherapy had no impact on survival. Conclusions Conservatively treated patients with severe AS have a grave prognosis, and it is worse in the presence of advanced age, LV dysfunction, heart failure, and renal failure.

Patients with asymptomatic severe aortic stenosis (AS) are reported to have a benign prognosis and hence the American College of Cardiology/American Heart Association guidelines do not recommend aortic valve replacement (AVR) for patients with isolated asymptomatic severe AS. However, symptoms are subjective and would depend upon patient's life style. We examined the natural and unnatural history of initially asymptomatic patients with severe AS.A search of our echocardiographic database between 1993 and 2003 yielded 740 patients with severe AS defined as aortic valve area 0.8 cm2 or less. Thorough chart reviews were conducted to collect clinical and pharmacologic data. Of these, 338 patients were asymptomatic at the initial encounter forming the study cohort.Patient characteristics were the following: age 71 +/- 15 years, males 51%, aortic valve area 0.72 +/- 0.17 cm2, left ventricular ejection fraction 0.59 +/- 0.17. Ninety-nine (29%) patients had AVR during a mean follow-up of 3.5 years. Survival at 1, 2, and 5 years in the nonoperated patients were 67%, 56%, and 38%, respectively, compared with 94%, 93%, and 90% in those who underwent AVR (p0.0001). The Cox regression model was used to adjust for the effect of 18 clinical, echocardiographic, and pharmacologic variables on survival. The adjusted hazard ratio for death with AVR was 0.17 (95% confidence interval [CI] 0.10 to 0.29). In the nonoperated group, renal insufficiency (risk ratio [RR] 3.1, 95% CI 1.5 to 6.6), beta blocker use (RR 0.52, 95% CI 0.31 to 0.88), statin use (RR 0.52, 95% CI 0.27 to 0.99), age (per year RR 1.03, 95% CI 1.02 to 1.05), and left ventricular ejection fraction (per % RR 0.99, 95% CI 0.98 to 1.00) were found to be the independent predictors of mortality. The benefit of AVR was further supported by sensitivity and propensity score analyses.Our observational data indicate that the natural history of asymptomatic AS is not benign and that survival is dramatically improved by AVR. Survival of the asymptomatic unoperated or nonoperable patients may potentially be improved by the use of beta blockers and statins.

Background: Calcific aortic stenosis (AS) is a disease of the elderly. However, there is reluctance to offer aortic valve replacement (AVR) for elderly patients with severe AS. We investigated if AVR confers a survival benefit in elderly patients with severe AS. Methods: We screened our echocardiographic database from 1993 to 2003 for patients with severe AS (AVarea 0.8 cm 2 ) and age 80 years. Two hundred and seventy seven patients were identified. Complete chart reviews were performed for clinical data. Mortality data were obtained from National Death Index. Survival curves of patients who underwent AVR during the follow-up period were compared with those managed nonsurgically. Results: Patient characteristics were as follows: age 85 4 years, 53% male, AVarea 0.68 0.16 cm 2 ,EF52 20%, CAD 47%, diabetes 17%. Over a mean follow-up of 2.5 years, 55 (20%) had AVR and there were 175deaths. One-year, 2-yearand 5-year survival ratesamong patients with AVR were 87, 78 and 68% respectively, compared with 52, 40 and 22%, respectively, in those who had no AVR (p < 0.0001). Hazard ratio for death with AVR adjusted for 19 covariates including age, EF, gender, comorbidities and pharmacotherapy was 0.38 (95% CI 0.26—0.66, p < 0.0001). Conclusion: Prognosis of medically managed severe calcific AS in the elderly patients is dismal. AVR appears to improve survival of these patients and should be strongly considered in the absence of other major comorbidities.

ObjectivesThis study evaluated the clinical outcomes of consecutive, selected patients treated with coronary artery bypass graft (CABG) surgery or percutaneous coronary intervention (PCI) with drug-eluting stents (DES) for unprotected left main coronary artery (ULMCA) disease.BackgroundAlthough recent data suggest that PCI with DES provides better clinical outcomes compared to bare-metal stenting for ULMCA disease, there is a paucity of data comparing PCI with DES to CABG.MethodsSince April 2003, when DES first became available at our institution, 123 patients underwent CABG, and 50 patients underwent PCI with DES for ULMCA disease.ResultsHigh-risk patients (Parsonnet score >15) comprised 46% of the CABG group and 64% of the PCI group (p = 0.04). The 30-day major adverse cardiac and cerebrovascular event (MACCE) rate for CABG and PCI was 17% and 2% (p < 0.01), respectively. The mean follow-up was 6.7 ± 6.2 months in the CABG group and 5.6 ± 3.9 months in the PCI group (p = 0.26). The estimated MACCE-free survival at six months and one year was 83% and 75% in the CABG group versus 89% and 83% in the PCI group (p = 0.20). By multivariable Cox regression, Parsonnet score, diabetes, and CABG were independent predictors of MACCE.ConclusionsDespite a higher percentage of high-risk patients, PCI with DES for ULMCA disease was not associated with an increase in immediate or medium-term complications compared with CABG. Our data suggest that a randomized comparison between the two revascularization strategies for ULMCA may be warranted.

Despite the use of embolic protection devices, no-reflow can still occur during saphenous vein grafts (SVGs) intervention. High-dose intracoronary adenosine infusion preconditions the myocardium, improves coronary flow, and prevents no-reflow. The role of high-dose intragraft adenosine infusion on protection of microvascular function and prevention of no-reflow has not been investigatedWe investigated the cardioprotective effect of high-dose intragraft adenosine infusion, compared with placebo, on microvascular function and prevention of no-reflow during SVGs intervention.We randomized 22 patients with SVGs stenoses to receive either a 10-min intragraft adenosine infusion (200 μg/min; total dose = 2,000 μg) or normal saline prior to stenting. Average peak velocity (APV), coronary flow velocity reserve (CVR), thrombolysis in myocardial infarction (TIMI) frame count (TFC), TIMI myocardial perfusion grade (TMPG), and the rate of no-reflow were compared between the two groups before adenosine or saline infusions and after stentingAfter stenting, hyperemic APV, CVR, and TMPG were significantly higher in the adenosine-treated group than in the control group (60 ± 18 vs. 35 ± 10 cm/sec; 2.6 ± 0.54 vs. 1.8 ± 0.47; and 2.8 ± 0.90 vs. 2.1 ± 0.80, respectively; P0.05. TFC was significantly lower in the adenosine-treated group than in the control group (14 ± 3.0 vs. 26 ± 13; P0.05). In the control group, four patients (36%) developed no-reflow compared to none in the adenosine-treated patient; P0.05 CONCLUSIONS: This study provides the first evidence that high-dose intragraft adenosine infusion compared with placebo protects microvascular function and prevents no-reflow during SVGs intervention.

OBJECTIVES We compared resting and hyperemic pressure gradients induced by intrarenal papaverine for the assessment of renal artery stenosis (RAS). We also investigated the incidence of the QT interval prolongation and ventricular arrhythmias. BACKGROUND In the coronary circulation, maximal hyperemia is essential in determining the significance of a stenosis. In the renal circulation, the role of maximal hyperemia for the assessment of RAS has not been established. METHODS In 55 patients with RAS (67 RAS), resting P(d)/P(a) ratio (the ratio between distal renal pressure to the aortic pressure), renal fractional flow reserve (FFR), and resting and hyperemic systolic gradients (RSG and HSG, respectively) were measured with a pressure guidewire. In a subset of 16 patients, renal vein renin activity (RVRA) was measured. RESULTS HSG was significantly greater than RSG (20 ± 14 mm Hg vs. 9.0 ± 13 mm Hg, respectively; P < 0.001). Renal FFR was significantly lower than baseline P(d)/P(a) ratio (0.91 ± 0.06 vs. 0.94 ± 0.06 vs. respectively; P < 0.001). RVRA increased from 50 ± 66% at rest to 122 ± 112% at hyperemia, P < 0.01. At HSG of 21 mm Hg or renal FFR of 0.90, RVRA increased markedly (120%), but RVRA increased modestly (18%) when RSG was 16 mm Hg or resting P(d)/P(a) ratio was 0.93. The corrected QT intervals at baseline vs. hyperemia were not significantly different (433 ± 26 vs. 436 ± 25 msec, respectively; P = NS); no episodes of ventricular arrhythmias were noted. CONCLUSIONS Renin production, an index of renal ischemia, was markedly greater at hyperemia than at rest, suggesting that RAS, with either an HSG of 21 mm Hg or a renal FFR of 0.90, can be considered a hemodynamically significant stenosis. Intrarenal papaverine neither prolonged the QT interval nor induced ventricular arrhythmias and the safety of which will need to be corroborated in a large study. © 2010 Wiley-Liss, Inc.

Secure Multi-Party Computation (SMC) allows parties with similar background to compute results upon their private data, minimizing the threat of disclosure. The exponential increase in sensitive data that needs to be passed upon networked computers and the stupendous growth of internet has precipitated vast opportunities for cooperative computation, where parties come together to facilitate computations and draw out conclusions that are mutually beneficial; at the same time aspiring to keep their private data secure. These computations are generally required to be done between competitors, who are obviously weary of each-others intentions. SMC caters not only to the needs of such parties but also provides plausible solutions to individual organizations for problems like privacy-preserving database query, privacy-preserving scientific computations, privacy-preserving intrusion detection and privacy-preserving data mining. This paper is the second version of a previously proposed protocol Extended_Encrypto_Random, which itself was an extension of the initial work Encrytpo_Random. The former endeavors presented a plain sailing yet effective approach to SMC and also laid forward an aptly crafted architecture, whereby such efficient protocols, involving the parties that have come forward for joint-computations and the third party who undertakes such computations, can be developed. Through this succeeding work an attempt has been made to further strengthen the existing protocols, thus paving the way for a more secure multi-party computational process.

Objectives: The aim of this study was to compare the procedural characteristics and outcomes of patients with acute myocardial infarction treated with drug-eluting stents (DES) vs. bare metal stents (BMS). Background: DES have been shown to reduce the incidence of restenosis and target vessel revascularization (TVR) in clinical randomized studies when compared with BMS in patients undergoing elective percutaneous intervention. Limited data are available with the use of DES in patients with acute ST-segment elevation myocardial infarction. Methods: Two hundred and sixty-one consecutive patients who presented with myocardial infarction between 7/2001 and 8/2005 were studied. The procedural characteristics, 30-day and 12-month outcomes of 131 patients treated with DES were compared with 130 patients treated with BMS. Results: At 12-months follow-up DES therapy was associated with a substantial decrease in major adverse cardiovascular events (MACE) (HR 0.33; P =0.002), TVR (HR 0.19; P =0.002), and recurrent myocardial infarction (HR 0.23; P =0.051) vs. BMS therapy. Coronary interventions utilizing DES were characterized by a marked increase in the number of stent per target vessel (DES: 1.9 ± 0.9 vs. BMS: 1.38 ± 0.6, P < 0.0001), treatment of bifurcation (DES: 21% vs. BMS: 5%, P =0.0004), and multivessel intervention (DES: 22% vs. BMS: 8%, P =0.003). Conclusion: The routine use of DES in acute myocardial infarction is associated with reduced rates of MACE at 12 months vs BMS, despite a higher rate of complex procedures in the DES treated patients. In addition to its anti-restenosis effect, the improved outcome of patients treated with DES may be linked to a more complete revascularization in association with prolonged clopidogrel therapy. © 2007 Wiley-Liss, Inc.

Background Pulmonary hypertension complicating severe aortic stenosis increases morbidity and mortality. Causes and mechanisms of this are unclear. Methods This is a retrospective observational study of 626 patients with severe aortic stenosis who had measurable pulmonary arterial pressure by Doppler echocardiography. Clinical, echocardiographic and pharmacological data were related to the presence of pulmonary hypertension. Results Of the 626 patients, 119 (19%) had severe pulmonary hypertension defined as pulmonary artery systolic pressure ≥60 mmHg. Patients with severe pulmonary hypertension had a smaller aortic valve area ( P < 0.0001), a lower left ventricular ejection fraction ( P < 0.0001), a higher mitral E/A velocity ratio ( P < 0.0001) indicating a higher filling pressure and a higher prevalence of 3 or 4+ mitral regurgitation ( P < 0.001). They were less likely to be on a beta blocker ( P = 0.05) or a statin ( P = 0.02). Smaller aortic valve area, left ventricular dysfunction, mitral regurgitation and lack of statin use were independent predictors of severe pulmonary hypertension. Conclusions Severity of aortic stenosis, left ventricular dysfunction, and mitral regurgitation are risk factors for the genesis of pulmonary hypertension and statins may potentially be protective in patients with severe aortic stenosis.

Background. Severe pulmonary arterial hypertension in patients with severe aortic stenosis (AS) carries a poor prognosis. There are limited data on the effect of aortic valve replacement (AVR) in these patients. Methods. Our echocardiographic database between 1993 and 2003 was searched for patients with severe AS defined as a Doppler estimated aortic valve area of 0.8 cm 2 or less and severe pulmonary hypertension defined as a pulmonary arterial systolic pressure 60 mm Hg or greater. Of the 740 patients with severe AS, 119 (16%) had severe pulmonary hypertension forming the study cohort. The AVR was performed in 36 (30%) of these patients. Survival of patients with and without AVR were compared and adjusted for comorbidities and group differences using the Cox regression model. Results. Characteristics of patients with severe pulmonary hypertension; age 75 13 years, 39% women, left ventricular ejection fraction 41 20%. Patients who underwent AVR had a significantly higher five-year survival of 65% compared with 20% for those treated medically (p < 0.0001). The relative mortality risk associated with AVR was 0.28 (95% confidence interval 0.22 to 0.36) and was independent of age, gender, ejection fraction, diabetes, coronary disease, serum creatinine level, and concomitant medical therapy such as beta blockers, angiotensin converting inhibitors, and statins. The benefit of AVR was further supported by sensitivity and propensity score analyses. Patients on conservative therapy had a 30-day mortality of 30% and a one-year mortality of 70%. Conclusions. Aortic valve replacement in patients with severe pulmonary hypertension secondary to severe AS is associated with a huge survival benefit. Medical therapy alone carries a dismal prognosis and AVR should be considered urgently in severe AS patients with severe pulmonary hypertension. (Ann Thorac Surg 2007;84:80‐6) © 2007 by The Society of Thoracic Surgeons P ulmonary hypertension in patients with severe aortic stenosis (AS) is associated with the presence of left ventricular (LV) dysfunction, mitral regurgitation, and higher LV filling pressures [1‐4]. In small observational series it has been shown to be associated with high mortality in the absence of aortic valve replacement (AVR). In a series of 47 patients with severe pulmonary hypertension associated with severe AS, 37 patients undergoing AVR had a survival five-year rate of 52% compared with 20% for the 10 patients treated medically [5]. Smaller observational series have reported a potential symptomatic and hemodynamic benefit with AVR in such patients [4, 6, 7]. In view of the paucity of data on outcomes in these patients, we analyzed the effect of AVR in a larger cohort of patients with severe AS and severe pulmonary hypertension from our institution. We constructed a comprehensive database including all clinical comorbidities and pharmacologic therapy, along with echocardiographic and surgical details.

Janus is the Roman god of gates and doors, of beginnings and endings, represented as a double-faced head looking in opposite directions. In this issue of the Journal , Cho et al. ([1][1]) lead us to look both forward and backward as they compare the effect of a 6-day course of granulocyte colony