Accurate and reliable Magnetic Resonance Imaging (MRI) analysis is particularly important for adaptive radiotherapy, a recent medical advance capable of improving cancer diagnosis and treatment. Recent studies have shown that IVIM-NET, a deep neural network (DNN), can achieve high accuracy in MRI analysis, indicating the potential of deep learning to enhance diagnostic capabilities in healthcare. However, IVIM-NET does not provide calibrated uncertainty information needed for reliable and trustworthy predictions in healthcare. Moreover, the expensive computation and memory demands of IVIM-NET reduce hardware performance, hindering widespread adoption in realistic scenarios. To address these challenges, this paper proposes an algorithm-hardware co-optimization flow for high-performance and reliable MRI analysis. At the algorithm level, a transformation design flow is introduced to convert IVIM-NET to a mask-based Bayesian Neural Network (BayesNN), facilitating reliable and efficient uncertainty estimation. At the hardware level, we propose an FPGA-based accelerator with several hardware optimizations, such as mask-zero skipping and operation reordering. Experimental results demonstrate that our co-design approach can satisfy the uncertainty requirements of MRI analysis, while achieving 7.5 times and 32.5 times speedup on an Xilinx VU13P FPGA compared to GPU and CPU implementations with reduced power consumption.

Human analysis of cardiac MRI scans is subject to enough noise and bias that a quick machine-learning approach easily matched it for accuracy, researchers found. A single expert reader contouring a scan to get left ventricular (LV) ejection fraction and LV mass had intra-observer error manifest as coefficients of variation of 5.4 and 3.8, respectively, while junior trainees had similar 5.2 and 5.5 coefficients. When a scan was repeated on the same person but at a different time, there was no difference in overall variation when comparing results from an expert, two trainees, and an automated deep-learning neural network, reported Charlotte Manisty, PhD, of University College London and Barts Heart Centre, and colleagues in Circulation: Cardiovascular Imaging. "Given that the greatest sources of measurement error were human factors (i.e., non-modifiable intra- and inter-observer variability), we believe that, with improvement, it is only a matter of time before automated approaches are super-human," according to the investigators. "These data demonstrated that human (intra-observer) error was greater than half of scan-rescan error, an effect that was not minimized by an expert when compared with junior clinicians after appropriate training, despite fifteen years' additional experience," they added.

Leesburg, VA, April 2, 2018 - A test of machine-learning algorithms shows promise for computer-aided prognosis of acute spinal cord injury, according to a study to be presented at the ARRS 2018 Annual Meeting, set for April 22-27 in Washington, DC. The study to be presented by Jason Talbot, assistant professor of radiology at the University of California, San Francisco, involved using semiautomated image analysis with machine-learning algorithms to assess the accuracy of axial T2-weighted radiomic features for classifying patients by degree of neurologic injury. Several machine-learning algorithms were tested for injury classification based on texture variables. For each trained model, the accuracy of predicting the testing set was recorded, as were variables important to the model. This proof-of-principle study highlights the feasibility of applying a semiautomated MRI analysis pipeline for atlas-based texture feature extraction from T2-weighted MRI at the epicenter of acute spinal cord injury (SCI).