1
|
Winzeck S, Mocking SJT, Bezerra R, Bouts MJRJ, McIntosh EC, Diwan I, Garg P, Chutinet A, Kimberly WT, Copen WA, Schaefer PW, Ay H, Singhal AB, Kamnitsas K, Glocker B, Sorensen AG, Wu O. Ensemble of Convolutional Neural Networks Improves Automated Segmentation of Acute Ischemic Lesions Using Multiparametric Diffusion-Weighted MRI. AJNR Am J Neuroradiol 2019; 40:938-945. [PMID: 31147354 DOI: 10.3174/ajnr.a6077] [Citation(s) in RCA: 29] [Impact Index Per Article: 5.8] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 10/01/2018] [Accepted: 04/19/2019] [Indexed: 11/07/2022]
Abstract
BACKGROUND AND PURPOSE Accurate automated infarct segmentation is needed for acute ischemic stroke studies relying on infarct volumes as an imaging phenotype or biomarker that require large numbers of subjects. This study investigated whether an ensemble of convolutional neural networks trained on multiparametric DWI maps outperforms single networks trained on solo DWI parametric maps. MATERIALS AND METHODS Convolutional neural networks were trained on combinations of DWI, ADC, and low b-value-weighted images from 116 subjects. The performances of the networks (measured by the Dice score, sensitivity, and precision) were compared with one another and with ensembles of 5 networks. To assess the generalizability of the approach, we applied the best-performing model to an independent Evaluation Cohort of 151 subjects. Agreement between manual and automated segmentations for identifying patients with large lesion volumes was calculated across multiple thresholds (21, 31, 51, and 70 cm3). RESULTS An ensemble of convolutional neural networks trained on DWI, ADC, and low b-value-weighted images produced the most accurate acute infarct segmentation over individual networks (P < .001). Automated volumes correlated with manually measured volumes (Spearman ρ = 0.91, P < .001) for the independent cohort. For the task of identifying patients with large lesion volumes, agreement between manual outlines and automated outlines was high (Cohen κ, 0.86-0.90; P < .001). CONCLUSIONS Acute infarcts are more accurately segmented using ensembles of convolutional neural networks trained with multiparametric maps than by using a single model trained with a solo map. Automated lesion segmentation has high agreement with manual techniques for identifying patients with large lesion volumes.
Collapse
Affiliation(s)
- S Winzeck
- From the Department of Radiology (S.W., S.J.T.M., R.B., M.J.R.J.B., E.C.M., I.D., P.G., H.A., A.G.S., O.W.), Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, Massachusetts.,Division of Anaesthesia (S.W.), Department of Medicine, University of Cambridge, Cambridge, UK
| | - S J T Mocking
- From the Department of Radiology (S.W., S.J.T.M., R.B., M.J.R.J.B., E.C.M., I.D., P.G., H.A., A.G.S., O.W.), Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, Massachusetts
| | - R Bezerra
- From the Department of Radiology (S.W., S.J.T.M., R.B., M.J.R.J.B., E.C.M., I.D., P.G., H.A., A.G.S., O.W.), Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, Massachusetts
| | - M J R J Bouts
- From the Department of Radiology (S.W., S.J.T.M., R.B., M.J.R.J.B., E.C.M., I.D., P.G., H.A., A.G.S., O.W.), Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, Massachusetts
| | - E C McIntosh
- From the Department of Radiology (S.W., S.J.T.M., R.B., M.J.R.J.B., E.C.M., I.D., P.G., H.A., A.G.S., O.W.), Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, Massachusetts
| | - I Diwan
- From the Department of Radiology (S.W., S.J.T.M., R.B., M.J.R.J.B., E.C.M., I.D., P.G., H.A., A.G.S., O.W.), Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, Massachusetts
| | - P Garg
- From the Department of Radiology (S.W., S.J.T.M., R.B., M.J.R.J.B., E.C.M., I.D., P.G., H.A., A.G.S., O.W.), Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, Massachusetts
| | - A Chutinet
- Departments of Neurology (A.C., W.T.K., H.A., A.B.S.).,Department of Medicine (A.C.), Faculty of Medicine, Chulalongkorn University, King Chulalongkorn Memorial Hospital, Thai Red Cross Society, Bangkok, Thailand
| | - W T Kimberly
- Departments of Neurology (A.C., W.T.K., H.A., A.B.S.)
| | - W A Copen
- Radiology (W.A.C., P.W.S.), Massachusetts General Hospital, Boston, Massachusetts
| | - P W Schaefer
- Radiology (W.A.C., P.W.S.), Massachusetts General Hospital, Boston, Massachusetts
| | - H Ay
- From the Department of Radiology (S.W., S.J.T.M., R.B., M.J.R.J.B., E.C.M., I.D., P.G., H.A., A.G.S., O.W.), Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, Massachusetts.,Departments of Neurology (A.C., W.T.K., H.A., A.B.S.)
| | - A B Singhal
- Departments of Neurology (A.C., W.T.K., H.A., A.B.S.)
| | - K Kamnitsas
- Department of Computing (K.K., B.G.), Imperial College London, London, UK
| | - B Glocker
- Department of Computing (K.K., B.G.), Imperial College London, London, UK
| | - A G Sorensen
- From the Department of Radiology (S.W., S.J.T.M., R.B., M.J.R.J.B., E.C.M., I.D., P.G., H.A., A.G.S., O.W.), Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, Massachusetts
| | - O Wu
- From the Department of Radiology (S.W., S.J.T.M., R.B., M.J.R.J.B., E.C.M., I.D., P.G., H.A., A.G.S., O.W.), Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, Massachusetts
| |
Collapse
|
2
|
Viana PÃCC, Horvat N, Guglielmetti G, Coelho R, Nahas WC, Park R, Bezerra R, Bastos DA, Rodrigues T, Vargas HA. The accuracy of multiparametric magnetic resonance imaging (mpMRI) using PI-RADS v2 in disease upgrading on re-biopsy among patients with low-risk prostate cancer (PCa) on active surveillance (AS): A Brazilian perspective. J Clin Oncol 2017. [DOI: 10.1200/jco.2017.35.15_suppl.5084] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 11/20/2022] Open
Abstract
5084 Background: The current selection criteria to AS is critical and it becomes even more relevant in Latin America, given the higher proportion of high risk cancers. The objective of this study is to analyze the accuracy of mpMRI using PI-RADS v2 in predicting the risk of upgrading on re-biopsy (UR) in men with low-risk PCa on AS. Methods: In this Institutional Review Board approved prospective study, patients with low-grade PCa selected for AS at our institution underwent mpMRI at least 6 weeks after the baseline 12-core random prostate biopsy (BSB), from March 2014 to March 2016. One blinded abdominal radiologist evaluated the exams regarding presence of dominant lesion and assigned the PI-RADS v2 score. MRI-target transrectal ultrasound-guided re-biopsies were performed in all patients within 6-12 months after the BSB. Standardized 12-core biopsy was performed and additional cores were taken from suspicious areas on mpMRI. Results: One hundred and nine patients were included, 93 (85.3%) patients had a dominant lesion on MRI. mpMRI were classified as PI-RADS 1, 2 or 3 in 67 (61.5%) patients, and as PI-RADS 4 or 5 in 42 (38.5%) patients. UR occurred in 42 (38.5%) patients. Out of these, 39 (92.8%) had radical prostatectomy, 6 (15.4%) T2a, 24 (61.5%) T2b, and 9 (23.1%) T3a. The diagnostic performance of mpMRI for PCa upgrading after re-biopsy was summarized in table 1. Patients assigned as PI-RADS 4 or 5 presented a significantly higher risk for UR compared with patients with PI-RADS 1, 2 or 3 (73.8% vs 16.4%, p < 0.001). Logistic regression analyses demonstrated that PI-RADS 4 or 5 remained a significant predictor of UR (OR: 37.366, p < 0.0001). Conclusions: We demonstrated in our population that mpMRI using PI-RADS v2 is a significant predictor for upgrading on re-biopsy in patients on AS and could be used to guide TRUS biopsy, increasing the accuracy of current clinical criteria for AS. [Table: see text]
Collapse
Affiliation(s)
| | | | | | - Rafael Coelho
- Sao Paulo Cancer Institute ICESP - University of Sao Paulo FMUSP, São Paulo, Brazil
| | | | - Rubens Park
- Instituto do Cancer do Estado de São Paulo, São Paulo, Brazil
| | | | - Diogo Assed Bastos
- Sao Paulo State Cancer Institute - University of Sao Paulo, São Paulo, Brazil
| | | | | |
Collapse
|
6
|
Fridman S, Bezerra R, Cagy M, Basile L, Piedade RA, Ribeiro P. [The effects of bromazepam on contingent negative variation and reaction time in a visuomotor task]. Rev Neurol 2006; 43:398-402. [PMID: 17006858] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [MESH Headings] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 05/12/2023]
Abstract
INTRODUCTION Bromazepam is the second most commonly used benzodiazepine in Brazil. Psychophysiological research on this substance is still in its early stages. AIM To determine the neurotoxicity of bromazepam by examining reaction times (RT) and contingent negative variations (CNV). SUBJECTS AND METHODS Using a videogame produced in our laboratory for psychophysiological research purposes (Car Acquisition), 14 healthy volunteers (9 males) aged between 23 and 42 drove a vehicle along a road full of curves (i.e. distractors) while they had to respond to imperative stimuli (i.e. orders to press the button on the joystick) that were preceded by warnings (S1-S2-RM paradigm with distractor). We compared RT, amplitudes and latencies of the CNV at each of the three electrodes on the median line (Fz, Cz and Pz) one hour after random, double-blind and crossed administration of placebo (P), 3 mg of bromazepam (B3) or 6 mg of bromazepam (B6) on different days. STATISTICS one-way ANOVA and Post Hoc Scheffé. RESULTS No significant differences were observed in the RT. At Pz, the CNV amplitudes displayed significant differences for P, B3 and B6 (p = 0.006), and also for B3 and B6 (p = 0.018), with B6 > B3 = P. At Fz, a non-significant tendency (p = 0.074) suggested a difference between the latencies, shorter in B6 than in B3 (p = 0.098), both equivalent to placebo. The mean amplitudes ranged between 2.4 and 5.9 microV. CONCLUSIONS Behavioural and neurophysiological neurotoxicity was insignificant one hour after administration of a single 3 or 6 mg dose of bromazepam in healthy young adults. Low mean amplitudes were compatible with the interference from distractors and did not result in floor effect.
Collapse
Affiliation(s)
- S Fridman
- Instituto de Psiquiatría, Universidade Federal do Rio de Janeiro, Rio de Janeiro, Brasil.
| | | | | | | | | | | |
Collapse
|