Patient-specific modeling for guided rehabilitation of stroke patients: the BrainX3 use-case

Summary Background Stroke is a leading cause of mortality and disability worldwide and the economic costs of treatment and post-stroke care are substantial. The Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) provides a systematic, comparable method of quantifying health loss by disease, age, sex, year, and location to provide information to health systems and policy makers on more than 300 causes of disease and injury, including stroke. The results presented here are the estimates of burden due to overall stroke and ischaemic and haemorrhagic stroke from GBD 2016. Methods We report estimates and corresponding uncertainty intervals (UIs), from 1990 to 2016, for incidence, prevalence, deaths, years of life lost (YLLs), years lived with disability (YLDs), and disability-adjusted life-years (DALYs). DALYs were generated by summing YLLs and YLDs. Cause-specific mortality was estimated using an ensemble modelling process with vital registration and verbal autopsy data as inputs. Non-fatal estimates were generated using Bayesian meta-regression incorporating data from registries, scientific literature, administrative records, and surveys. The Socio-demographic Index (SDI), a summary indicator generated using educational attainment, lagged distributed income, and total fertility rate, was used to group countries into quintiles. Findings In 2016, there were 5·5 million (95% UI 5·3 to 5·7) deaths and 116·4 million (111·4 to 121·4) DALYs due to stroke. The global age-standardised mortality rate decreased by 36·2% (−39·3 to −33·6) from 1990 to 2016, with decreases in all SDI quintiles. Over the same period, the global age-standardised DALY rate declined by 34·2% (−37·2 to −31·5), also with decreases in all SDI quintiles. There were 13·7 million (12·7 to 14·7) new stroke cases in 2016. Global age-standardised incidence declined by 8·1% (−10·7 to −5·5) from 1990 to 2016 and decreased in all SDI quintiles except the middle SDI group. There were 80·1 million (74·1 to 86·3) prevalent cases of stroke globally in 2016; 41·1 million (38·0 to 44·3) in women and 39·0 million (36·1 to 42·1) in men. Interpretation Although age-standardised mortality rates have decreased sharply from 1990 to 2016, the decrease in age-standardised incidence has been less steep, indicating that the burden of stroke is likely to remain high. Planned updates to future GBD iterations include generating separate estimates for subarachnoid haemorrhage and intracerebral haemorrhage, generating estimates of transient ischaemic attack, and including atrial fibrillation as a risk factor. Funding Bill & Melinda Gates Foundation

We propose a dual pathway, 11-layers deep, three-dimensional Convolutional Neural Network for the challenging task of brain lesion segmentation. The devised architecture is the result of an in-depth analysis of the limitations of current networks proposed for similar applications. To overcome the computational burden of processing 3D medical scans, we have devised an efficient and effective dense training scheme which joins the processing of adjacent image patches into one pass through the network while automatically adapting to the inherent class imbalance present in the data. Further, we analyze the development of deeper, thus more discriminative 3D CNNs. In order to incorporate both local and larger contextual information, we employ a dual pathway architecture that processes the input images at multiple scales simultaneously. For post-processing of the network's soft segmentation, we use a 3D fully connected Conditional Random Field which effectively removes false positives. Our pipeline is extensively evaluated on three challenging tasks of lesion segmentation in multi-channel MRI patient data with traumatic brain injuries, brain tumours, and ischemic stroke. We improve on the state-of-the-art for all three applications, with top ranking performance on the public benchmarks BRATS 2015 and ISLES 2015. Our method is computationally efficient, which allows its adoption in a variety of research and clinical settings. The source code of our implementation is made publicly available.