scholarly journals Three-Dimensional Convolutional Autoencoder Extracts Features of Structural Brain Images With a “Diagnostic Label-Free” Approach: Application to Schizophrenia Datasets

2021 ◽  
Vol 15 ◽  
Author(s):  
Hiroyuki Yamaguchi ◽  
Yuki Hashimoto ◽  
Genichi Sugihara ◽  
Jun Miyata ◽  
Toshiya Murai ◽  
...  
Keyword(s):  
Psychiatric Disorders ◽  
Brain Imaging ◽  
Spatial Information ◽  
Three Dimensional ◽  
Clinical Information ◽  
Label Free ◽  
Brain Images ◽  
Convolutional Autoencoder ◽  
Mri Scans ◽  
Magnetic Resonance Imaging Mri

There has been increasing interest in performing psychiatric brain imaging studies using deep learning. However, most studies in this field disregard three-dimensional (3D) spatial information and targeted disease discrimination, without considering the genetic and clinical heterogeneity of psychiatric disorders. The purpose of this study was to investigate the efficacy of a 3D convolutional autoencoder (3D-CAE) for extracting features related to psychiatric disorders without diagnostic labels. The network was trained using a Kyoto University dataset including 82 patients with schizophrenia (SZ) and 90 healthy subjects (HS) and was evaluated using Center for Biomedical Research Excellence (COBRE) datasets, including 71 SZ patients and 71 HS. We created 16 3D-CAE models with different channels and convolutions to explore the effective range of hyperparameters for psychiatric brain imaging. The number of blocks containing two convolutional layers and one pooling layer was set, ranging from 1 block to 4 blocks. The number of channels in the extraction layer varied from 1, 4, 16, and 32 channels. The proposed 3D-CAEs were successfully reproduced into 3D structural magnetic resonance imaging (MRI) scans with sufficiently low errors. In addition, the features extracted using 3D-CAE retained the relation to clinical information. We explored the appropriate hyperparameter range of 3D-CAE, and it was suggested that a model with 3 blocks may be related to extracting features for predicting the dose of medication and symptom severity in schizophrenia.

scholarly journals Three-dimensional convolutional autoencoder extracts features of structural brain images with a “diagnostic label-free” approach: Application to schizophrenia datasets

2020 ◽  
Author(s):  
Hiroyuki Yamaguchi ◽  
Yuki Hashimoto ◽  
Genichi Sugihara ◽  
Jun Miyata ◽  
Toshiya Murai ◽  
...  
Keyword(s):  
Psychiatric Disorders ◽  
Spatial Information ◽  
Three Dimensional ◽  
Clinical Information ◽  
Label Free ◽  
Brain Images ◽  
Diagnostic Labels ◽  
Convolutional Autoencoder ◽  
Mri Scans ◽  
Magnetic Resonance Imaging Mri

ABSTRACTThere has been increasing interest in performing psychiatric brain imaging studies using deep learning. However, most studies in this field disregard three-dimensional (3D) spatial information and targeted disease discrimination, without considering the genetic and clinical heterogeneity of psychiatric disorders. The purpose of this study was to investigate the efficacy of a 3D convolutional autoencoder (CAE) for extracting features related to psychiatric disorders without diagnostic labels. The network was trained using a Kyoto University dataset including 82 patients with schizophrenia (SZ) and 90 healthy subjects (HS), and was evaluated using Center for Biomedical Research Excellence (COBRE) datasets including 71 SZ patients and 71 HS. The proposed 3D-CAEs were successfully reconstructed into high-resolution 3D structural magnetic resonance imaging (MRI) scans with sufficiently low errors. In addition, the features extracted using 3D-CAE retained the relevant clinical information. We explored the appropriate hyper parameter range of 3D-CAE, and it was suggested that a model with eight convolution layers might be relevant to extract features for predicting the dose of medication and symptom severity in schizophrenia.

scholarly journals Ultrasonic image analysis and image-guided interventions

2011 ◽  
Vol 1 (4) ◽  
pp. 673-685 ◽  
Author(s):  
J. Alison Noble ◽  
Nassir Navab ◽  
H. Becher
Keyword(s):  
Image Analysis ◽  
Imaging Modality ◽  
Medical Image Analysis ◽  
Three Dimensional ◽  
Clinical Information ◽  
Image Guided ◽  
Software Algorithms ◽  
Positron Emission ◽  
Imaging Research ◽  
Magnetic Resonance Imaging Mri

The fields of medical image analysis and computer-aided interventions deal with reducing the large volume of digital images (X-ray, computed tomography, magnetic resonance imaging (MRI), positron emission tomography and ultrasound (US)) to more meaningful clinical information using software algorithms. US is a core imaging modality employed in these areas, both in its own right and used in conjunction with the other imaging modalities. It is receiving increased interest owing to the recent introduction of three-dimensional US, significant improvements in US image quality, and better understanding of how to design algorithms which exploit the unique strengths and properties of this real-time imaging modality. This article reviews the current state of art in US image analysis and its application in image-guided interventions. The article concludes by giving a perspective from clinical cardiology which is one of the most advanced areas of clinical application of US image analysis and describing some probable future trends in this important area of ultrasonic imaging research.

scholarly journals 3D Segmentation of Perivascular Spaces on T1-Weighted 3 Tesla MR Images With a Convolutional Autoencoder and a U-Shaped Neural Network

2021 ◽  
Vol 15 ◽  
Author(s):  
Philippe Boutinaud ◽  
Ami Tsuchida ◽  
Alexandre Laurent ◽  
Filipa Adonias ◽  
Zahra Hanifehlou ◽  
...  
Keyword(s):  
Learning Performance ◽  
Similar Distribution ◽  
Perivascular Spaces ◽  
Visual Rating ◽  
Independent Expert ◽  
Convolutional Autoencoder ◽  
Mri Scans ◽  
Experienced Operator ◽  
Magnetic Resonance Imaging Mri ◽  
The Individual

We implemented a deep learning (DL) algorithm for the 3-dimensional segmentation of perivascular spaces (PVSs) in deep white matter (DWM) and basal ganglia (BG). This algorithm is based on an autoencoder and a U-shaped network (U-net), and was trained and tested using T1-weighted magnetic resonance imaging (MRI) data from a large database of 1,832 healthy young adults. An important feature of this approach is the ability to learn from relatively sparse data, which gives the present algorithm a major advantage over other DL algorithms. Here, we trained the algorithm with 40 T1-weighted MRI datasets in which all “visible” PVSs were manually annotated by an experienced operator. After learning, performance was assessed using another set of 10 MRI scans from the same database in which PVSs were also traced by the same operator and were checked by consensus with another experienced operator. The Sorensen-Dice coefficients for PVS voxel detection in DWM (resp. BG) were 0.51 (resp. 0.66), and 0.64 (resp. 0.71) for PVS cluster detection (volume threshold of 0.5 within a range of 0 to 1). Dice values above 0.90 could be reached for detecting PVSs larger than 10 mm3 and 0.95 for PVSs larger than 15 mm3. We then applied the trained algorithm to the rest of the database (1,782 individuals). The individual PVS load provided by the algorithm showed a high agreement with a semi-quantitative visual rating done by an independent expert rater, both for DWM and for BG. Finally, we applied the trained algorithm to an age-matched sample from another MRI database acquired using a different scanner. We obtained a very similar distribution of PVS load, demonstrating the interoperability of this algorithm.

Human Vocal Tract Modeling and Geometric Parameterization

2004 ◽  
Author(s):  
Hussain Z. Tameem ◽  
Bhavin V. Mehta
Keyword(s):  
Vocal Tract ◽  
Three Dimensional ◽  
Parametric Modeling ◽  
Three Dimensional Model ◽  
Non Invasive ◽  
Solid Models ◽  
Mri Scans ◽  
Invasive Method ◽  
Geometrical Information ◽  
Magnetic Resonance Imaging Mri

This investigation uses a multi disciplinary approach to standardize a non-invasive method for measuring human vocal tract morphology. A series of Magnetic Resonance Imaging (MRI) scans are performed on the subject’s vocal tract and a detailed three-dimensional model is created through image processing and computer modeling. This information is compared with the vocal tract measurements obtained with Eccovision Acoustic Pharyngometer, in order to establish the accuracy of the instrument. The model is then used to develop other specific models through parametric modeling. This method is useful in creating solid models with limited geometrical information and helps researchers study the human vocal tract changes due to aging and degenerative diseases.

Association of Inter-individual Differences in Imaging Markers with Schizophrenia Phenotypes

2017 ◽  
Vol 41 (S1) ◽  
pp. S43-S44
Author(s):  
G. Pergola ◽  
T. Quarto ◽  
M. Papalino ◽  
P. Di Carlo ◽  
P. Selvaggi ◽  
...  
Keyword(s):  
Brain Imaging ◽  
Clinical Course ◽  
Clinical Information ◽  
Roc Curves ◽  
Individual Variability ◽  
Clinical Population ◽  
Imaging Data ◽  
Mri Scans ◽  
Brain Imaging Data ◽  
Genetically Determined

IntroductionNeuroimaging studies have identified several candidate biomarkers of schizophrenia. However, it is unclear whether the considerable variability in these neurobiological correlates between patients can be translated into the clinical setting.ObjectivesWe aimed to identify neuroimaging predictors of clinical course in patients with schizophrenia. Combined with the identification of genetically determined markers of schizophrenia risk, our studies aimed to elucidate the biological basis and the clinical relevance of inter-individual variability between patients.MethodsWe included over 150 patients with schizophrenia and 279 healthy volunteers across five neuroimaging centers in the framework of the IMAGEMEND project [1]. We performed multiple studies on MRI scans using random forests and ROC curves to predict clinical course. Data from healthy controls served to normalize the data from the clinical population and to provide a benchmark for the findings.ResultsWe identified ensembles of neuroimaging markers and of genetic variants predictive of clinical course. Results highlight that (i) brain imaging carries significant clinical information, (ii) clinical information at baseline can considerably increase prediction accuracy.ConclusionThe methodological challenges and the results will be discussed in the context of recent findings from other multi-site studies. We conclude that brain imaging data on their own right are relevant to stratify patients in terms of clinical course; however, complementing these data with other modalities such as genetics and clinical information is necessary to further develop the field towards clinical application of the predictions.Disclosure of interestGiulio Pergola is the academic supervisor of a Hoffmann-La Roche Collaboration grant that partially funds his salary.

scholarly journals Isotropically resolved label-free tomographic imaging based on tomographic moulds for optical trapping

2021 ◽  
Vol 10 (1) ◽  
Author(s):  
Moosung Lee ◽  
Kyoohyun Kim ◽  
Jeonghun Oh ◽  
YongKeun Park
Keyword(s):  
Optical Tweezers ◽  
Blood Cells ◽  
Spatial Information ◽  
Three Dimensional ◽  
Label Free ◽  
Light Traps ◽  
Axial Resolution ◽  
Isotropic Resolution ◽  
3D Microscopy ◽  
Structural Details

AbstractA major challenge in three-dimensional (3D) microscopy is to obtain accurate spatial information while simultaneously keeping the microscopic samples in their native states. In conventional 3D microscopy, axial resolution is inferior to spatial resolution due to the inaccessibility to side scattering signals. In this study, we demonstrate the isotropic microtomography of free-floating samples by optically rotating a sample. Contrary to previous approaches using optical tweezers with multiple foci which are only applicable to simple shapes, we exploited 3D structured light traps that can stably rotate freestanding complex-shaped microscopic specimens, and side scattering information is measured at various sample orientations to achieve isotropic resolution. The proposed method yields an isotropic resolution of 230 nm and captures structural details of colloidal multimers and live red blood cells, which are inaccessible using conventional tomographic microscopy. We envision that the proposed approach can be deployed for solving diverse imaging problems that are beyond the examples shown here.

scholarly journals Semi-Automated and Direct Localization and Labeling of EEG Electrodes Using MR Structural Images for Simultaneous fMRI-EEG

2020 ◽  
Vol 14 ◽  
Author(s):  
Abhishek S. Bhutada ◽  
Pradyumna Sepúlveda ◽  
Rafael Torres ◽  
Tomás Ossandón ◽  
Sergio Ruiz ◽  
...  
Keyword(s):  
Spatial Information ◽  
Three Dimensional ◽  
Spatial Localization ◽  
Accurate Identification ◽  
Brain Scans ◽  
3D Space ◽  
Automated Method ◽  
Simultaneous Acquisition ◽  
Additional Equipment ◽  
Magnetic Resonance Imaging Mri

Electroencephalography (EEG) source reconstruction estimates spatial information from the brain’s electrical activity acquired using EEG. This method requires accurate identification of the EEG electrodes in a three-dimensional (3D) space and involves spatial localization and labeling of EEG electrodes. Here, we propose a new approach to tackle this two-step problem based on the simultaneous acquisition of EEG and magnetic resonance imaging (MRI). For the step of spatial localization of electrodes, we extract the electrode coordinates from the curvature of the protrusions formed in the high-resolution T1-weighted brain scans. In the next step, we assign labels to each electrode based on the distinguishing feature of the electrode’s distance profile in relation to other electrodes. We then compare the subject’s electrode data with template-based models of prelabeled distance profiles of correctly labeled subjects. Based on this approach, we could localize EEG electrodes in 26 head models with over 90% accuracy in the 3D localization of electrodes. Next, we performed electrode labeling of the subjects’ data with progressive improvements in accuracy: with ∼58% accuracy based on a single EEG-template, with ∼71% accuracy based on 3 EEG-templates, and with ∼76% accuracy using 5 EEG-templates. The proposed semi-automated method provides a simple alternative for the rapid localization and labeling of electrodes without the requirement of any additional equipment than what is already used in an EEG-fMRI setup.

scholarly journals Comparison of Intraoperative 3-Dimensional Fluoroscopy With Standard Computed Tomography for Stereotactic Frame Registration

2019 ◽  
Author(s):  
Terrance Peng ◽  
Daniel R Kramer ◽  
Morgan B Lee ◽  
Michael F Barbaro ◽  
Li Ding ◽  
...  
Keyword(s):  
Computed Tomography ◽  
Ct Scan ◽  
Three Dimensional ◽  
Ct Scanner ◽  
3 Dimensional ◽  
Lead Placement ◽  
Stereotactic Frame ◽  
Mri Scans ◽  
Marginal Means ◽  
Magnetic Resonance Imaging Mri

Abstract BACKGROUND Three-dimensional fluoroscopy via the O-arm (Medtronic, Dublin, Ireland) has been validated for intraoperative confirmation of successful lead placement in stereotactic electrode implantation. However, its role in registration and targeting has not yet been studied. After frame placement, many stereotactic neurosurgeons obtain a computed tomography (CT) scan and merge it with a preoperative magnetic resonance imaging (MRI) scan to generate planning coordinates; potential disadvantages of this practice include increased procedure time and limited scanner availability. OBJECTIVE To evaluate whether the second-generation O-arm (O2) can be used in lieu of a traditional CT scan to obtain accurate frame-registration scans. METHODS In 7 patients, a postframe placement CT scan was merged with preoperative MRI and used to generate lead implantation coordinates. After implantation, the fiducial box was again placed on the patient to obtain an O2 confirmation scan. Vector, scalar, and Euclidean differences between analogous X, Y, and Z coordinates from fused O2/MRI and CT/MRI scans were calculated for 33 electrode target coordinates across 7 patients. RESULTS Marginal means of difference for vector (X = −0.079 ± 0.099 mm; Y = −0.076 ± 0.134 mm; Z = −0.267 ± 0.318 mm), scalar (X = −0.146 ± 0.160 mm; Y = −0.306 ± 0.106 mm; Z = 0.339 ± 0.407 mm), and Euclidean differences (0.886 ± 0.190 mm) remained within the predefined equivalence margin differences of −2 mm and 2 mm. CONCLUSION This study demonstrates that O2 may emerge as a viable alternative to the traditional CT scanner for generating planning coordinates. Adopting the O2 as a perioperative tool may offer reduced transport risks, decreased anesthesia time, and greater surgical efficiency.

scholarly journals Boarfish (Capros aper) target strength modelled from magnetic resonance imaging (MRI) scans of its swimbladder

2013 ◽  
Vol 70 (7) ◽  
pp. 1451-1459 ◽  
Author(s):  
Sascha M. M. Fässler ◽  
Ciaran O'Donnell ◽  
J. M. Jech
Keyword(s):  
Magnetic Resonance Imaging ◽  
Magnetic Resonance ◽  
Three Dimensional ◽  
Target Strength ◽  
Fish Length ◽  
Commercial Fishing ◽  
Resonance Imaging ◽  
Length Relationship ◽  
Mri Scans ◽  
Magnetic Resonance Imaging Mri

Abstract Fässler, S. M. M., O'Donnell, C., and Jech, J.M. 2013. Boarfish (Capros aper) target strength modelled from magnetic resonance imaging (MRI) scans of its swimbladder. – ICES Journal of Marine Science, 70: . Boarfish (Capros aper) abundance has increased dramatically in the Northeast Atlantic from the early 1970s after successive years of good recruitment attributed to an increase in sea surface temperature. Due to increased commercial fishing over recent years, an acoustic boarfish survey funded by the Killybegs Fishermen's Organisation was initiated by the Marine Institute to establish a baseline for the future management of this stock. In the absence of any species-specific boarfish target strength (TS), acoustic backscatter was estimated by a Kirchhoff-ray mode model using reconstructed three-dimensional swimbladder shapes which were computed from magnetic resonance imaging scans of whole fish. The model predicted TS as a function of size, fish tilt angle, and operating frequency. Standardized directivity patterns revealed the increasing importance of changes in the inclination of the dorsal swimbladder surface at higher frequencies (120 and 200 kHz) and a less directive response at lower frequencies (18 and 38 kHz). The model predicted a TS-to-total fish length relationship of TS = 20 log10(L) − 66.2. The intercept is ∼1 dB higher than in the general physoclist relationship, potentially reflecting the bulky nature of the boarfish swimbladder with its relatively large circumference.

scholarly journals RatLesNetv2: A Fully Convolutional Network for Rodent Brain Lesion Segmentation

2020 ◽  
Vol 14 ◽  
Author(s):  
Juan Miguel Valverde ◽  
Artem Shatillo ◽  
Riccardo De Feo ◽  
Olli Gröhn ◽  
Alejandra Sierra ◽  
...  
Keyword(s):  
Focal Cerebral Ischemia ◽  
Brain Mri ◽  
Brain Lesion ◽  
Three Dimensional ◽  
Medical Image Segmentation ◽  
Lesion Segmentation ◽  
Convolutional Network ◽  
Brain Images ◽  
Mri Scans ◽  
Mr Brain Images

We present a fully convolutional neural network (ConvNet), named RatLesNetv2, for segmenting lesions in rodent magnetic resonance (MR) brain images. RatLesNetv2 architecture resembles an autoencoder and it incorporates residual blocks that facilitate its optimization. RatLesNetv2 is trained end to end on three-dimensional images and it requires no preprocessing. We evaluated RatLesNetv2 on an exceptionally large dataset composed of 916 T2-weighted rat brain MRI scans of 671 rats at nine different lesion stages that were used to study focal cerebral ischemia for drug development. In addition, we compared its performance with three other ConvNets specifically designed for medical image segmentation. RatLesNetv2 obtained similar to higher Dice coefficient values than the other ConvNets and it produced much more realistic and compact segmentations with notably fewer holes and lower Hausdorff distance. The Dice scores of RatLesNetv2 segmentations also exceeded inter-rater agreement of manual segmentations. In conclusion, RatLesNetv2 could be used for automated lesion segmentation, reducing human workload and improving reproducibility. RatLesNetv2 is publicly available at https://github.com/jmlipman/RatLesNetv2.

Sign in / Sign up

Export Citation Format

Share Document