Person:
Pomerantz, Stuart

Profile Picture

Email Address

AA Acceptance Date

Birth Date

Research Projects

Organizational Units

Job Title

Last Name

Pomerantz

First Name

Stuart

Name

Pomerantz, Stuart
Author's Publications: search.filters.isAuthorOfPublication.9353db7b-8f36-4b93-8fad-f8b3e150fa19

Search Results

Now showing 1 - 3 of 3
  • Thumbnail Image
    Publication
    Optimal Brain MRI Protocol for New Neurological Complaint
    (Public Library of Science, 2014) Mehan, William; González, R. Gilberto; Buchbinder, Bradley; Chen, John; Copen, William; Gupta, Rajiv; Hirsch, Joshua; Hunter, George; Hunter, Scott; Johnson, Jason M.; Kelly, Hillary R.; Larvie, Mykol; Lev, Michael; Pomerantz, Stuart; Rapalino, Otto; Rincon, Sandra; Romero, Javier; Schaefer, Pamela; Shah, Vinil
    Background/Purpose Patients with neurologic complaints are imaged with MRI protocols that may include many pulse sequences. It has not been documented which sequences are essential. We assessed the diagnostic accuracy of a limited number of sequences in patients with new neurologic complaints. Methods: 996 consecutive brain MRI studies from patients with new neurological complaints were divided into 2 groups. In group 1, reviewers used a 3-sequence set that included sagittal T1-weighted, axial T2-weighted fluid-attenuated inversion recovery, and axial diffusion-weighted images. Subsequently, another group of studies were reviewed using axial susceptibility-weighted images in addition to the 3 sequences. The reference standard was the study's official report. Discrepancies between the limited sequence review and the reference standard including Level I findings (that may require immediate change in patient management) were identified. Results: There were 84 major findings in 497 studies in group 1 with 21 not identified in the limited sequence evaluations: 12 enhancing lesions and 3 vascular abnormalities identified on MR angiography. The 3-sequence set did not reveal microhemorrhagic foci in 15 of 19 studies. There were 117 major findings in 499 studies in group 2 with 19 not identified on the 4-sequence set: 17 enhancing lesions and 2 vascular lesions identified on angiography. All 87 Level I findings were identified using limited sequence (56 acute infarcts, 16 hemorrhages, and 15 mass lesions). Conclusion: A 4-pulse sequence brain MRI study is sufficient to evaluate patients with a new neurological complaint except when contrast or angiography is indicated.
  • Thumbnail Image
    Publication
    The Massachusetts General Hospital Acute Stroke Imaging Algorithm: An Experience and Evidence Based Approach
    (BMJ Publishing Group, 2013) Gonzalez, Ramon; Copen, William; Schaefer, Pamela; Lev, Michael; Pomerantz, Stuart; Rapalino, Otto; Chen, John; Hunter, George; Romero, Javier; Buchbinder, Bradley; Larvie, Mykol; Hirsch, Joshua; Gupta, Rajiv
    The Massachusetts General Hospital Neuroradiology Division employed an experience and evidence based approach to develop a neuroimaging algorithm to best select patients with severe ischemic strokes caused by anterior circulation occlusions (ACOs) for intravenous tissue plasminogen activator and endovascular treatment. Methods found to be of value included the National Institutes of Health Stroke Scale (NIHSS), non-contrast CT, CT angiography (CTA) and diffusion MRI. Perfusion imaging by CT and MRI were found to be unnecessary for safe and effective triage of patients with severe ACOs. An algorithm was adopted that includes: non-contrast CT to identify hemorrhage and large hypodensity followed by CTA to identify the ACO; diffusion MRI to estimate the core infarct; and NIHSS in conjunction with diffusion data to estimate the clinical penumbra.
  • No Thumbnail Available
    Publication
    An explainable deep-learning algorithm for the detection of acute intracranial haemorrhage from small datasets
    (Springer Science and Business Media LLC, 2018-12-17) Lee, Hyunkwang; Yune, Sehyo; Mansouri, Mohammad; Kim, Myeongchan; Tajmir, Shahein H.; Guerrier, Claude E.; Ebert, Sarah A.; Pomerantz, Stuart; Romero, Javier; Kamalian, Mohammad; Gonzalez, Ramon; Lev, Michael; Do, Synho
    Owing to improvements in image recognition via deep learning, machine-learning algorithms could eventually be applied to automated medical diagnoses that can guide clinical decision-making. However, these algorithms remain a 'black box' in terms of how they generate the predictions from the input data. Also, high-performance deep learning requires large, high-quality training datasets. Here, we report the development of an understandable deep-learning system that detects acute intracranial haemorrhage (ICH) and classifies five ICH subtypes from unenhanced head computed-tomography scans. By using a dataset of only 904 cases for algorithm training, the system achieved a performance similar to that of expert radiologists in two independent test datasets containing 200 cases (sensitivity of 98% and specificity of 95%) and 196 cases (sensitivity of 92% and specificity of 95%). The system includes an attention map and a prediction basis retrieved from training data to enhance explainability, and an iterative process that mimics the workflow of radiologists. Our approach to algorithm development can facilitate the development of deep-learning systems for a variety of clinical applications and accelerate their adoption into clinical practice.