Quantifying and leveraging predictive uncertainty for medical image assessment

The interpretation of medical images is a challenging task, often complicated by the presence of artifacts, occlusions, limited contrast and more. Most notable is the case of chest radiography, where there is a high inter-rater variability in the detection and classification of abnormalities. This is largely due to inconclusive evidence in the data or subjective definitions of disease appearance. An additional example is the classification of anatomical views based on 2D Ultrasound images. Often, the anatomical context captured in a frame is not sufficient to recognize the underlying anatomy. Current machine learning solutions for these problems are typically limited to providing probabilistic predictions, relying on the capacity of underlying models to adapt to limited information and the high degree of label noise. In practice, however, this leads to overconfident systems with poor generalization on unseen data. To account for this, we propose a system that learns not only the probabilistic estimate for classification, but also an explicit uncertainty measure which captures the confidence of the system in the predicted output. We argue that this approach is essential to account for the inherent ambiguity characteristic of medical images from different radiologic exams including computed radiography, ultrasonography and magnetic resonance imaging. In our experiments we demonstrate that sample rejection based on the predicted uncertainty can significantly improve the ROC-AUC for various tasks, e.g., by 8% to 0.91 with an expected rejection rate of under 25% for the classification of different abnormalities in chest radiographs. In addition, we show that using uncertainty-driven bootstrapping to filter the training data, one can achieve a significant increase in robustness and accuracy. Finally, we present a multi-reader study showing that the predictive uncertainty is indicative of reader errors.     

A comprehensive method for calculating total body irradiation

PurposeTo provide means for calculating the dose received by various tissues of the patient, calculate lung shield, and verify received dose using a phantom as a tool for quality assurance for a planned Total Body Irradiation (TBI) procedure in radiotherapy.MethodUsing Microsoft Visual Basic, MATLAB, and Python, a program for Total Body Irradiation Calculation in Radiotherapy (TBICR) is constructed. It uses patient translation and beam zone method for total body irradiation calculations to compute the proper dose received by the patient and determine the lung shield thickness. There are three main user-friendly interfaces in the application. The first one allows the user to upload the TBI topography and estimate the distances needed for TBI calculations. The second one enables the user to count the number of beam zones needed for each point and estimate the effective area (A_eff) for each level. The third interface estimates the velocity required to deliver the relative dose depending on patient separation, Monitor Units (MU), couch speed and travel distance. It allows the user to compute the required lung shield thickness, read any patient's CT DICOM file and acquire dose in any distinct location using machine learning model to predict the dose.ResultsThe TBICR software has been successfully validated by reproducing all of the manual calculations in an exact and timely manner. TBICR generated more accurate results and confirmed the absorbed dose to patient through measurements on Anderson phantom.ConclusionsA computer program for the calculation of total body irradiation (TBI) is described in full. The dose received at each point on the patient, the calculation of lung shield and the determination of the velocity and time required for the couch movement are all made possible using the software. The ease of use, precision, data storage and printing are some important features of the present software.     

Super-resolution Ultrasound Imaging

The majority of exchanges of oxygen and nutrients are performed around vessels smaller than 100 μm, allowing cells to thrive everywhere in the body. Pathologies such as cancer, diabetes and arteriosclerosis can profoundly alter the microvasculature. Unfortunately, medical imaging modalities only provide indirect observation at this scale. Inspired by optical microscopy, ultrasound localization microscopy has bypassed the classic compromise between penetration and resolution in ultrasonic imaging. By localization of individual injected microbubbles and tracking of their displacement with a subwavelength resolution, vascular and velocity maps can be produced at the scale of the micrometer. Super-resolution ultrasound has also been performed through signal fluctuations with the same type of contrast agents, or through switching on and off nano-sized phase-change contrast agents. These techniques are now being applied pre-clinically and clinically for imaging of the microvasculature of the brain, kidney, skin, tumors and lymph nodes.     

Predictors of postoperative acute kidney injury in patients undergoing hip fracture surgery: A systematic review and meta-analysis

BackgroundThe present study aimed to summarize the predictors of acute kidney injury (AKI) in patients after hip surgery.MethodsA literature search was performed using PubMed, EMBASE, Cochrane Library, and Web of Science for studies assessing the predictors of AKI after hip fracture surgery. Pooled odds ratio (OR) and mean difference (MD) of those who experienced AKI compared to those who did not were calculated for each variable. Evidence was assessed using the Newcastle–Ottawa Scale.ResultsTen studies with 34 potential factors were included in the meta-analysis. In the primary analysis, 12 factors were associated with AKI, comprising males (OR 1.25; 95% confidence interval (CI) 1.14–1.36), advanced age (MD 2.28; 95% CI 0.80–3.75), myocardial infarction (OR 1.39; 95% CI 1.18–1.63), hypertension (OR 1.46; 95% CI 1.13–1.89), diabetes (OR 1.84; 95% CI 1.40–2.42), chronic kidney disease (OR 3.66; 95% CI 2.21–6.07), hip arthroplasty (OR 1.35; 95% CI 1.22–1.50), angiotensin-converting enzyme inhibitors/angiotensin receptor blockers use (OR 2.28; 95% CI 1.68–3.08), more intraoperative blood loss (MD 44.06; 95% CI 2.88–85.24), higher preoperative blood urea nitrogen levels (MD 5.29; 95% CI 3.38–7.20), higher preoperative serum creatinine levels (MD 0.4; 95% CI 0.26–0.53), and lower preoperative estimated glomerular filtration rate (MD −19.59; 95% CI −26.92–−12.26). Another 13 factors related to AKI in individual studies were identified in the systematic review.ConclusionRelated prophylaxis strategies should be implemented in patients involved with the above-mentioned characteristics to prevent AKI after hip surgery.     

A Novel Multiparametric Score for the Detection and Grading of Prosthetic Mitral Valve Obstruction in Cases With Different Disc Motion Abnormalities

Prosthetic mechanical valves are the elective choice in mitral valve (MV) replacement, because of their reliability and easiness of implantation. However, these prostheses can suffer from complications, the major one being prosthetic mitral valve thrombosis (PMVT). In these cases, transthoracic doppler echocardiogram (TDE) is the standard diagnostic workup for diagnosis of valve malfunction. The American Society of Echocardiography (ASE) indicates the possible TDE-derived indexes, which can help in identifying insurgence of MV replacement complications. Unfortunately, in some cases, it is not possible to detect PMVT based on these criteria. In these cases, we speak of Doppler silent thrombosis and only more accurate and invasive analyses, such as fluoroscopy, allow for a correct diagnosis. In this work, computational fluid dynamic models were implemented to simulate valve fluid dynamics in different clinical scenarios in order to improve the reliability of PMVT diagnosis based on TDE. In detail, seven mechanical valve configurations, associated to different potential thrombotic conditions (symmetric and asymmetric stenosis), were designed and tested using five pathologic transmitral velocity profile, extracted from real TDE images; to obtain the flow rate profiles, each TDE velocity profile was scaled to yield a mean flow rate (MFR) of 4, 5 and 6 L/min, respectively. As a result, 105 (7 × 5 × 3) synthetic cases, accounting for different velocity profiles, MFRs and valve configurations, were simulated. TDE-derived indexes were calculated according to the ASE guidelines that were extracted. Advanced statistical methods were applied to propose a new diagnostic algorithm for detecting PMVT. Our results showed that there isn't any significant difference between symmetric and asymmetric stenosis, probe location and flow rate waveform and confirmed that the single modality diagnostic is not able to predict thrombosis in a relevant number of cases, referable to mild and mild-severe stenosis cases. To overcome the problem, a novel multi-parametric discrete score based on the designed diagnostic algorithm was attained and tested; the percentage of stenosis (POS) was predicted with an accuracy rate of 90.5%. Even more interestingly, the error rate of 9.5% is related to four false positive cases corresponding to mild stenosis (POS = 15%) which were erroneously classified as mild-severe stenosis. No false negatives were obtained. Our results suggest that a reliable estimation must take into account the mean flow rate as well as the transmitral velocity profile in order to provide a correct diagnosis.