| 1. |
|
|
| 2. |
- Ahlander, Britt-Marie, 1954-, et al.
(författare)
-
An echo-planar imaging sequence is superior to a steady-state free precession sequence for visual as well as quantitative assessment of cardiac magnetic resonance stress perfusion
- 2017
-
Ingår i: Clinical Physiology and Functional Imaging. - : Blackwell Publishing. - 1475-0961 .- 1475-097X. ; 37:1, s. 52-61
-
Tidskriftsartikel (refereegranskat)abstract
- BACKGROUND: To assess myocardial perfusion, steady-state free precession cardiac magnetic resonance (SSFP, CMR) was compared with gradient-echo-echo-planar imaging (GRE-EPI) using myocardial perfusion scintigraphy (MPS) as reference.METHODS: Cardiac magnetic resonance perfusion was recorded in 30 patients with SSFP and in another 30 patients with GRE-EPI. Timing and extent of inflow delay to the myocardium was visually assessed. Signal-to-noise (SNR) and contrast-to-noise (CNR) ratios were calculated. Myocardial scar was visualized with a phase-sensitive inversion recovery sequence (PSIR). All scar positive segments were considered pathologic. In MPS, stress and rest images were used as in clinical reporting. The CMR contrast wash-in slope was calculated and compared with the stress score from the MPS examination. CMR scar, CMR perfusion and MPS were assessed separately by one expert for each method who was blinded to other aspects of the study.RESULTS: Visual assessment of CMR had a sensitivity for the detection of an abnormal MPS at 78% (SSFP) versus 91% (GRE-EPI) and a specificity of 58% (SSFP) versus 84% (GRE-EPI). Kappa statistics for SSFP and MPS was 0·29, for GRE-EPI and MPS 0·72. The ANOVA of CMR perfusion slopes for all segments versus MPS score (four levels based on MPS) had correlation r = 0·64 (SSFP) and r = 0·96 (GRE-EPI). SNR was for normal segments 35·63 ± 11·80 (SSFP) and 17·98 ± 8·31 (GRE-EPI), while CNR was 28·79 ± 10·43 (SSFP) and 13·06 ± 7·61 (GRE-EPI).CONCLUSION: GRE-EPI displayed higher agreement with the MPS results than SSFP despite significantly lower signal intensity, SNR and CNR.
|
|
| 3. |
- Ali, Muhaddisa Barat, 1986, et al.
(författare)
-
A novel federated deep learning scheme for glioma and its subtype classification
- 2023
-
Ingår i: Frontiers in Neuroscience. - 1662-4548 .- 1662-453X. ; 17
-
Tidskriftsartikel (refereegranskat)abstract
- Background: Deep learning (DL) has shown promising results in molecular-based classification of glioma subtypes from MR images. DL requires a large number of training data for achieving good generalization performance. Since brain tumor datasets are usually small in size, combination of such datasets from different hospitals are needed. Data privacy issue from hospitals often poses a constraint on such a practice. Federated learning (FL) has gained much attention lately as it trains a central DL model without requiring data sharing from different hospitals. Method: We propose a novel 3D FL scheme for glioma and its molecular subtype classification. In the scheme, a slice-based DL classifier, EtFedDyn, is exploited which is an extension of FedDyn, with the key differences on using focal loss cost function to tackle severe class imbalances in the datasets, and on multi-stream network to exploit MRIs in different modalities. By combining EtFedDyn with domain mapping as the pre-processing and 3D scan-based post-processing, the proposed scheme makes 3D brain scan-based classification on datasets from different dataset owners. To examine whether the FL scheme could replace the central learning (CL) one, we then compare the classification performance between the proposed FL and the corresponding CL schemes. Furthermore, detailed empirical-based analysis were also conducted to exam the effect of using domain mapping, 3D scan-based post-processing, different cost functions and different FL schemes. Results: Experiments were done on two case studies: classification of glioma subtypes (IDH mutation and wild-type on TCGA and US datasets in case A) and glioma grades (high/low grade glioma HGG and LGG on MICCAI dataset in case B). The proposed FL scheme has obtained good performance on the test sets (85.46%, 75.56%) for IDH subtypes and (89.28%, 90.72%) for glioma LGG/HGG all averaged on five runs. Comparing with the corresponding CL scheme, the drop in test accuracy from the proposed FL scheme is small (−1.17%, −0.83%), indicating its good potential to replace the CL scheme. Furthermore, the empirically tests have shown that an increased classification test accuracy by applying: domain mapping (0.4%, 1.85%) in case A; focal loss function (1.66%, 3.25%) in case A and (1.19%, 1.85%) in case B; 3D post-processing (2.11%, 2.23%) in case A and (1.81%, 2.39%) in case B and EtFedDyn over FedAvg classifier (1.05%, 1.55%) in case A and (1.23%, 1.81%) in case B with fast convergence, which all contributed to the improvement of overall performance in the proposed FL scheme. Conclusion: The proposed FL scheme is shown to be effective in predicting glioma and its subtypes by using MR images from test sets, with great potential of replacing the conventional CL approaches for training deep networks. This could help hospitals to maintain their data privacy, while using a federated trained classifier with nearly similar performance as that from a centrally trained one. Further detailed experiments have shown that different parts in the proposed 3D FL scheme, such as domain mapping (make datasets more uniform) and post-processing (scan-based classification), are essential.
|
|
| 4. |
- Ge, Chenjie, 1991, et al.
(författare)
-
Enlarged Training Dataset by Pairwise GANs for Molecular-Based Brain Tumor Classification
- 2020
-
Ingår i: IEEE Access. - 2169-3536 .- 2169-3536. ; 8:1, s. 22560-22570
-
Tidskriftsartikel (refereegranskat)abstract
- This paper addresses issues of brain tumor subtype classification using Magnetic Resonance Images (MRIs) from different scanner modalities like T1 weighted, T1 weighted with contrast-enhanced, T2 weighted and FLAIR images. Currently most available glioma datasets are relatively moderate in size, and often accompanied with incomplete MRIs in different modalities. To tackle the commonly encountered problems of insufficiently large brain tumor datasets and incomplete modality of image for deep learning, we propose to add augmented brain MR images to enlarge the training dataset by employing a pairwise Generative Adversarial Network (GAN) model. The pairwise GAN is able to generate synthetic MRIs across different modalities. To achieve the patient-level diagnostic result, we propose a post-processing strategy to combine the slice-level glioma subtype classification results by majority voting. A two-stage course-to-fine training strategy is proposed to learn the glioma feature using GAN-augmented MRIs followed by real MRIs. To evaluate the effectiveness of the proposed scheme, experiments have been conducted on a brain tumor dataset for classifying glioma molecular subtypes: isocitrate dehydrogenase 1 (IDH1) mutation and IDH1 wild-type. Our results on the dataset have shown good performance (with test accuracy 88.82%). Comparisons with several state-of-the-art methods are also included.
|
|
| 5. |
- Pfeiffer, Christoph, 1989, et al.
(författare)
-
On-scalp MEG sensor localization using magnetic dipole-like coils: A method for highly accurate co-registration
- 2020
-
Ingår i: Neuroimage. - : Elsevier BV. - 1053-8119 .- 1095-9572. ; 212
-
Tidskriftsartikel (refereegranskat)abstract
- Source modelling in magnetoencephalography (MEG) requires precise co-registration of the sensor array and the anatomical structure of the measured individual's head. In conventional MEG, the positions and orientations of the sensors relative to each other are fixed and known beforehand, requiring only localization of the head relative to the sensor array. Since the sensors in on-scalp MEG are positioned on the scalp, locations of the individual sensors depend on the subject's head shape and size. The positions and orientations of on-scalp sensors must therefore be measured a every recording. This can be achieved by inverting conventional head localization, localizing the sensors relative to the head - rather than the other way around. In this study we present a practical method for localizing sensors using magnetic dipole-like coils attached to the subject's head. We implement and evaluate the method in a set of on-scalp MEG recordings using a 7-channel on-scalp MEG system based on high critical temperature superconducting quantum interference devices (high-T-c SQUIDs). The method allows individually localizing the sensor positions, orientations, and responsivities with high accuracy using only a short averaging time (<= 2 mm, < 3 degrees and < 3%, respectively, with 1-s averaging), enabling continuous sensor localization. Calibrating and jointly localizing the sensor array can further improve the accuracy of position and orientation (< 1 mm and < 1 degrees, respectively, with 1-s coil recordings). We demonstrate source localization of on-scalp recorded somatosensory evoked activity based on coregistration with our method. Equivalent current dipole fits of the evoked responses corresponded well (within 4.2 mm) with those based on a commercial, whole-head MEG system.
|
|
| 6. |
- Hagberg, Eva, et al.
(författare)
-
Semi-supervised learning with natural language processing for right ventricle classification in echocardiography—a scalable approach
- 2022
-
Ingår i: Computers in Biology and Medicine. - : Elsevier BV. - 0010-4825 .- 1879-0534. ; 143
-
Tidskriftsartikel (refereegranskat)abstract
- We created a deep learning model, trained on text classified by natural language processing (NLP), to assess right ventricular (RV) size and function from echocardiographic images. We included 12,684 examinations with corresponding written reports for text classification. After manual annotation of 1489 reports, we trained an NLP model to classify the remaining 10,651 reports. A view classifier was developed to select the 4-chamber or RV-focused view from an echocardiographic examination (n = 539). The final models were two image classification models trained on the predicted labels from the combined manual annotation and NLP models and the corresponding echocardiographic view to assess RV function (training set n = 11,008) and size (training set n = 9951. The text classifier identified impaired RV function with 99% sensitivity and 98% specificity and RV enlargement with 98% sensitivity and 98% specificity. The view classification model identified the 4-chamber view with 92% accuracy and the RV-focused view with 73% accuracy. The image classification models identified impaired RV function with 93% sensitivity and 72% specificity and an enlarged RV with 80% sensitivity and 85% specificity; agreement with the written reports was substantial (both κ = 0.65). Our findings show that models for automatic image assessment can be trained to classify RV size and function by using model-annotated data from written echocardiography reports. This pipeline for auto-annotation of the echocardiographic images, using a NLP model with medical reports as input, can be used to train an image-assessment model without manual annotation of images and enables fast and inexpensive expansion of the training dataset when needed. © 2022
|
|
| 7. |
- Borrelli, P., et al.
(författare)
-
AI-based detection of lung lesions in F-18 FDG PET-CT from lung cancer patients
- 2021
-
Ingår i: Ejnmmi Physics. - : Springer Science and Business Media LLC. - 2197-7364. ; 8:1
-
Tidskriftsartikel (refereegranskat)abstract
- Background[F-18]-fluorodeoxyglucose (FDG) positron emission tomography with computed tomography (PET-CT) is a well-established modality in the work-up of patients with suspected or confirmed diagnosis of lung cancer. Recent research efforts have focused on extracting theragnostic and textural information from manually indicated lung lesions. Both semi-automatic and fully automatic use of artificial intelligence (AI) to localise and classify FDG-avid foci has been demonstrated. To fully harness AI's usefulness, we have developed a method which both automatically detects abnormal lung lesions and calculates the total lesion glycolysis (TLG) on FDG PET-CT.MethodsOne hundred twelve patients (59 females and 53 males) who underwent FDG PET-CT due to suspected or for the management of known lung cancer were studied retrospectively. These patients were divided into a training group (59%; n = 66), a validation group (20.5%; n = 23) and a test group (20.5%; n = 23). A nuclear medicine physician manually segmented abnormal lung lesions with increased FDG-uptake in all PET-CT studies. The AI-based method was trained to segment the lesions based on the manual segmentations. TLG was then calculated from manual and AI-based measurements, respectively and analysed with Bland-Altman plots.ResultsThe AI-tool's performance in detecting lesions had a sensitivity of 90%. One small lesion was missed in two patients, respectively, where both had a larger lesion which was correctly detected. The positive and negative predictive values were 88% and 100%, respectively. The correlation between manual and AI TLG measurements was strong (R-2 = 0.74). Bias was 42 g and 95% limits of agreement ranged from -736 to 819 g. Agreement was particularly high in smaller lesions.ConclusionsThe AI-based method is suitable for the detection of lung lesions and automatic calculation of TLG in small- to medium-sized tumours. In a clinical setting, it will have an added value due to its capability to sort out negative examinations resulting in prioritised and focused care on patients with potentially malignant lesions.
|
|
| 8. |
- Ge, Chenjie, 1991, et al.
(författare)
-
Multiscale Deep Convolutional Networks for Characterization and Detection of Alzheimer's Disease using MR Images
- 2019
-
Ingår i: Proceedings - International Conference on Image Processing, ICIP. - 1522-4880. ; 2019-September, s. 789-793
-
Konferensbidrag (refereegranskat)abstract
- This paper addresses the issues of Alzheimer's disease (AD) characterization and detection from Magnetic Resonance Images (MRIs). Many existing AD detection methods use single-scale feature learning from brain scans. In this paper, we propose a multiscale deep learning architecture for learning AD features. The main contributions of the paper include: (a) propose a novel 3D multiscale CNN architecture for the dedicated task of AD detection; (b) propose a feature fusion and enhancement strategy for multiscale features; (c) empirical study on the impact of several settings, including two dataset partitioning approaches, and the use of multiscale and feature enhancement. Experiments were conducted on an open ADNI dataset (1198 brain scans from 337 subjects), test results have shown the effectiveness of the proposed method with test accuracy of 93.53%, 87.24% (best, average) on subject separated dataset, and 99.44%, 98.80% (best, average) on random brain scan-partitioned dataset. Comparison with eight existing methods has provided further support to the proposed method.
|
|
| 9. |
- Reza, Salim, 1985-, et al.
(författare)
-
Smart dosimetry by pattern recognition using a single photon counting detector system in time over threshold mode
- 2012
-
Ingår i: Journal of Instrumentation. - 1748-0221 .- 1748-0221. ; 7:1
-
Tidskriftsartikel (refereegranskat)abstract
- The function of a dosimeter is to determine the absorbed dose of radiation, for those cases in which, generally, the particular type of radiation is already known. Lately, a number of applications have emerged in which all kinds of radiation are absorbed and are sorted by pattern recognition, such as the Medipix2 application in [1]. This form of smart dosimetry enables measurements where not only the total dosage is measured, but also the contributions of different types of radiation impacting upon the detector surface. Furthermore, the use of a photon counting system, where the energy deposition can be measured in each individual pixel, ensures measurements with a high degree of accuracy in relation to the pattern recognition. In this article a Timepix [2] detector system has been used in the creation of a smart dosimeter for Alpha, Beta and Gamma radiation. When a radioactive particle hits the detector surface it generates charge clusters and those impacting upon the detector surface are read out and image processing algorithms are then used to classify each charge cluster. The individual clusters are calculated and as a result, the dosage for each type of radiation is given. In some cases, several particles can impact in roughly the same place, forming overlapping clusters. In order to handle this problem, a cluster separation method has been added to the pattern recognition algorithm. When the clusters have been separated, they are classified by shape and sorted into the correct type of radiation. The algorithms and methods used in this dosimeter have been developed so as to be simple and computationally effective, in order to enable implementation on a portable device. © 2012 IOP Publishing Ltd and SISSA.
|
|
| 10. |
- Borrelli, P., et al.
(författare)
-
Artificial intelligence-aided CT segmentation for body composition analysis: a validation study
- 2021
-
Ingår i: European Radiology Experimental. - : Springer Science and Business Media LLC. - 2509-9280. ; 5:1
-
Tidskriftsartikel (refereegranskat)abstract
- BackgroundBody composition is associated with survival outcome in oncological patients, but it is not routinely calculated. Manual segmentation of subcutaneous adipose tissue (SAT) and muscle is time-consuming and therefore limited to a single CT slice. Our goal was to develop an artificial-intelligence (AI)-based method for automated quantification of three-dimensional SAT and muscle volumes from CT images.MethodsEthical approvals from Gothenburg and Lund Universities were obtained. Convolutional neural networks were trained to segment SAT and muscle using manual segmentations on CT images from a training group of 50 patients. The method was applied to a separate test group of 74 cancer patients, who had two CT studies each with a median interval between the studies of 3days. Manual segmentations in a single CT slice were used for comparison. The accuracy was measured as overlap between the automated and manual segmentations.ResultsThe accuracy of the AI method was 0.96 for SAT and 0.94 for muscle. The average differences in volumes were significantly lower than the corresponding differences in areas in a single CT slice: 1.8% versus 5.0% (p <0.001) for SAT and 1.9% versus 3.9% (p < 0.001) for muscle. The 95% confidence intervals for predicted volumes in an individual subject from the corresponding single CT slice areas were in the order of 20%.Conclusions The AI-based tool for quantification of SAT and muscle volumes showed high accuracy and reproducibility and provided a body composition analysis that is more relevant than manual analysis of a single CT slice.
|
|
