| 1. |
- 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.
|
|
| 2. |
- 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.
|
|
| 3. |
- Gryska, Emilia, 1992, et al.
(författare)
-
Deep learning for automatic brain tumour segmentation on MRI: evaluation of recommended reporting criteria via a reproduction and replication study.
- 2022
-
Ingår i: BMJ open. - : BMJ. - 2044-6055. ; 12:7
-
Tidskriftsartikel (refereegranskat)abstract
- To determine the reproducibility and replicability of studies that develop and validate segmentation methods for brain tumours on MRI and that follow established reproducibility criteria; and to evaluate whether the reporting guidelines are sufficient.Two eligible validation studies of distinct deep learning (DL) methods were identified. We implemented the methods using published information and retraced the reported validation steps. We evaluated to what extent the description of the methods enabled reproduction of the results. We further attempted to replicate reported findings on a clinical set of images acquired at our institute consisting of high-grade and low-grade glioma (HGG, LGG), and meningioma (MNG) cases.We successfully reproduced one of the two tumour segmentation methods. Insufficient description of the preprocessing pipeline and our inability to replicate the pipeline resulted in failure to reproduce the second method. The replication of the first method showed promising results in terms of Dice similarity coefficient (DSC) and sensitivity (Sen) on HGG cases (DSC=0.77, Sen=0.88) and LGG cases (DSC=0.73, Sen=0.83), however, poorer performance was observed for MNG cases (DSC=0.61, Sen=0.71). Preprocessing errors were identified that contributed to low quantitative scores in some cases.Established reproducibility criteria do not sufficiently emphasise description of the preprocessing pipeline. Discrepancies in preprocessing as a result of insufficient reporting are likely to influence segmentation outcomes and hinder clinical utilisation. A detailed description of the whole processing chain, including preprocessing, is thus necessary to obtain stronger evidence of the generalisability of DL-based brain tumour segmentation methods and to facilitate translation of the methods into clinical practice.
|
|
| 4. |
|
|
| 5. |
- Chiotis, K., et al.
(författare)
-
Longitudinal changes of tau PET imaging in relation to hypometabolism in prodromal and Alzheimer's disease dementia
- 2018
-
Ingår i: Molecular Psychiatry. - : Springer Science and Business Media LLC. - 1359-4184 .- 1476-5578. ; 23:7, s. 1666-1673
-
Tidskriftsartikel (refereegranskat)abstract
- The development of tau-specific positron emission tomography (PET) tracers allows imaging in vivo the regional load of tau pathology in Alzheimer's disease (AD) and other tauopathies. Eighteen patients with baseline investigations enroled in a 17-month follow-up study, including 16 with AD (10 had mild cognitive impairment and a positive amyloid PET scan, that is, prodromal AD, and six had AD dementia) and two with corticobasal syndrome. All patients underwent PET scans with [F-18]THK5317 (tau deposition) and [F-18]FDG (glucose metabolism) at baseline and follow-up, neuropsychological assessment at baseline and follow-up and a scan with [C-11]PIB (amyloid-beta deposition) at baseline only. At a group level, patients with AD (prodromal or dementia) showed unchanged [F-18]THK5317 retention over time, in contrast to significant decreases in [F-18]FDG uptake in temporoparietal areas. The pattern of changes in [F-18]THK5317 retention was heterogeneous across all patients, with qualitative differences both between the two AD groups (prodromal and dementia) and among individual patients. High [F-18]THK5317 retention was significantly associated over time with low episodic memory encoding scores, while low [F-18]FDG uptake was significantly associated over time with both low global cognition and episodic memory encoding scores. Both patients with corticobasal syndrome had a negative [C-11]PIB scan, high [F-18]THK5317 retention with a different regional distribution from that in AD, and a homogeneous pattern of increased [F-18]THK5317 retention in the basal ganglia over time. These findings highlight the heterogeneous propagation of tau pathology among patients with symptomatic AD, in contrast to the homogeneous changes seen in glucose metabolism, which better tracked clinical progression.
|
|
| 6. |
- Dalmo, Johanna, et al.
(författare)
-
Potential renal toxicity biomarkers indicating radiation injury after 177Lu-octreotate treatment
- 2013
-
Ingår i: Annual congress of the European association of nuclear medicine, october 19-23, 2013, Lyon, France. Posterwalk.
-
Konferensbidrag (övrigt vetenskapligt/konstnärligt)abstract
- The kidneys are one of the most exposed non-tumor tissues and regarded as one of the main dose-limiting organs in peptide receptor radionuclide therapy (PRRT). [177Lu-DOTA0, Tyr3]-octreotate (177Lu-octreotate) has shown promising results in the treatment of somatostatin receptor overexpressing neuroendocrine tumors, but optimization is still needed. The ability to give each patient as much 177Lu-octreotate as possible without inducing nephrotoxicity is necessary for an efficient treatment. However, due to large inter-individual differences in uptake and retention in the kidneys, there is a need for efficient Methods that early can indicate renal injury. A possible way is to identify biomarkers for high risk of radiation nephrotoxicity. The aim of this study was to investigate the potential of using urinary retinol binding protein (RBP), and blood valinhydantoin (VH) as biomarkers of nephrotoxicity on adult mice after 177Lu-octreotate treatment. BALB/c nude mice (n=6/group) were i.v. injected with 60 MBq or 120 MBq of 177Lu-octreotate. The control group was mock treated with saline. Spot urine samples were collected before injection, and 14, 30, 60 and 90 days after injection. Analysis of RBP4 and creatinine was performed using Mouse RBP4 ELISA kit and Creatinine kit from R&D Systems, respectively. Erythrocytes were separated from whole blood samples collected 90 days after injection, and analysed for VH by LC-MS/MS. The ratio between VH and a volumetric standard was calculated. The RBP/creatinine level increased with time in both groups given 177Lu-octreotate, with earlier and higher response for the 120 MBq group. No clear change in VH level between the different groups was observed. The result show that RBP may be a promising new biomarker for radiation induced kidney toxicity. The presently used method based on VH was not sensitive enough to be used as kidney toxicity marker. Further studies on mice are ongoing to validate if RBP4 may be efficient in predicting late nephrotoxicity. In patients, RBP/creatinine levels are followed in urine samples after treatment with 177Lu-octreotate.
|
|
| 7. |
- Ali, Muhaddisa Barat, 1986, et al.
(författare)
-
Multi-stream Convolutional Autoencoder and 2D Generative Adversarial Network for Glioma Classification
- 2019
-
Ingår i: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). - Cham : Springer International Publishing. - 1611-3349 .- 0302-9743. ; 11678 LNCS, s. 234-245
-
Konferensbidrag (refereegranskat)abstract
- Diagnosis and timely treatment play an important role in preventing brain tumor growth. Deep learning methods have gained much attention lately. Obtaining a large amount of annotated medical data remains a challenging issue. Furthermore, high dimensional features of brain images could lead to over-fitting. In this paper, we address the above issues. Firstly, we propose an architecture for Generative Adversarial Networks to generate good quality synthetic 2D MRIs from multi-modality MRIs (T1 contrast-enhanced, T2, FLAIR). Secondly, we propose a deep learning scheme based on 3-streams of Convolutional Autoencoders (CAEs) followed by sensor information fusion. The rational behind using CAEs is that it may improve glioma classification performance (as comparing with conventional CNNs), since CAEs offer noise robustness and also efficient feature reduction hence possibly reduce the over-fitting. A two-round training strategy is also applied by pre-training on GAN augmented synthetic MRIs followed by refined-training on original MRIs. Experiments on BraTS 2017 dataset have demonstrated the effectiveness of the proposed scheme (test accuracy 92.04%). Comparison with several exiting schemes has provided further support to the proposed scheme.
|
|
| 8. |
- Sundlöv, Anna, et al.
(författare)
-
Individualised Lu-177-DOTATATE treatment of neuroendocrine tumours based on kidney dosimetry
- 2017
-
Ingår i: European Journal of Nuclear Medicine and Molecular Imaging. - : Springer Science and Business Media LLC. - 1619-7070 .- 1619-7089. ; 44:9, s. 1480-1489
-
Tidskriftsartikel (refereegranskat)abstract
- Purpose To present data from an interim analysis of a Phase II trial designed to determine the feasibility, safety, and efficacy of individualising treatment based on renal dosimetry, by giving as many cycles as possible within a maximum renal biologically effective dose (BED). Method Treatment was given with repeated cycles of 7.4 GBq 177Lu-DOTATATE at 8-12-week intervals. Detailed dosimetry was performed in all patients after each cycle using a hybrid method (SPECT + planar imaging). All patients received treatment up to a renal BED of 27 +/- 2 Gy (alpha/beta = 2.6 Gy) (Step 1). Selected patients were offered further treatment up to a renal BED of 40 +/- 2 Gy (Step 2). Renal function was followed by estimation and measurement of the glomerular filtration rate (GFR). Results Fifty-one patients were included in the present analysis. Among the patients who received treatment as planned, the median number of cycles in Step 1 was 5 (range 3-7), and for those who completed Step 2 it was 7 (range 5-8); 73% were able to receive >4 cycles. Although GFR decreased in most patients after the completion of treatment, no grade 3-4 toxicity was observed. Patients with a reduced baseline GFR seemed to have an increased risk of GFR decline. Five patients received treatment in Step 2, none of whom exhibited a significant reduction in renal function. Conclusions Individualising PRRT using renal dosimetry seems feasible and safe and leads to an increased number of cycles in the majority of patients. The trial will continue as planned.
|
|
| 9. |
- Hagmarker, Linn, et al.
(författare)
-
Bone Marrow Absorbed Doses and Correlations with Hematologic Response During Lu-177-DOTATATE Treatments Are Influenced by Image-Based Dosimetry Method and Presence of Skeletal Metastases
- 2019
-
Ingår i: Journal of Nuclear Medicine. - : Society of Nuclear Medicine. - 0161-5505 .- 2159-662X. ; 60:10, s. 1406-1413
-
Tidskriftsartikel (refereegranskat)abstract
- This study aimed to compare different image-based methods for bone marrow dosimetry and study the dose-response relationship during treatment with Lu-177-DOTATATE in patients with and without skeletal metastases. Methods: This study included 46 patients with advanced neuroendocrine tumors treated with at least 2 fractions of Lu-177-DOTATATE at Sahlgrenska University Hospital. High- and low-uptake compartments were automatically outlined in planar images collected at 2, 24, 48, and 168 h after injection. The bone marrow absorbed doses were calculated from the cross doses of the high- and low-uptake compartments and the self-dose, using the time-activity concentration curve for the low-uptake compartment. This time-activity concentration curve was adjusted using a fixed constant of 1.8 for the planar dosimetry method and using the activity concentrations in vertebral bodies in SPECT images at 24 h after injection of Lu-177-DOTATATE in 4 hybrid methods: L4-SPECT used the activity concentration in the L4 vertebra, whereas V-SPECT, L-SPECT, and T-SPECT used the median activity concentration in all visible vertebrae, lumbar vertebrae, and thoracic vertebrae, respectively. Results: Using the planar method, L4-SPECT, V-SPECT, L-SPECT, and T-SPECT, the estimated median bone marrow absorbed doses were 0.19, 0.36, 0.40, 0.39, and 0.46 Gy/7.4 GBq, respectively, with respective ranges of 0.12-0.33, 0.15-1.44, 0.19-1.71, 0.21-1.60, and 0.18-2.12 Gy/7.4 GBq. For all methods, the bone marrow absorbed dose significantly correlated with decreased platelet counts. This correlation increased after treatment fraction 2: the Spearman correlation (r(s)) were -0.49 for the planar method, -0.61 for L4-SPECT, -0.63 for V-SPECT, -0.63 for L-SPECT, and -0.57 for T-SPECT. A separate analysis revealed an increased correlation for patients without skeletal metastases using the planar method (r(s) = -0.67). In contrast, hybrid methods had poor correlations for patients without metastases and stronger correlations for patients with skeletal metastases (r(s) = -0.61 to -0.74). The mean bone marrow absorbed doses were 3%-69% higher for patients with skeletal metastases than for patients without. Conclusion: The estimated bone marrow absorbed doses by image-based techniques and the correlation with platelets are influenced by the choice of measured vertebrae and the presence of skeletal metastases.
|
|
| 10. |
- Brabec, Jan, et al.
(författare)
-
Histogram analysis of tensor-valued diffusion MRI in meningiomas : Relation to consistency, histological grade and type
- 2022
-
Ingår i: NeuroImage: Clinical. - : Elsevier BV. - 2213-1582. ; 33
-
Tidskriftsartikel (refereegranskat)abstract
- BACKGROUND: Preoperative radiological assessment of meningioma characteristics is of value for pre- and post-operative patient management, counselling, and surgical approach.PURPOSE: To investigate whether tensor-valued diffusion MRI can add to the preoperative prediction of meningioma consistency, grade and type.MATERIALS AND METHODS: 30 patients with intracranial meningiomas (22 WHO grade I, 8 WHO grade II) underwent MRI prior to surgery. Diffusion MRI was performed with linear and spherical b-tensors with b-values up to 2000 s/mm2. The data were used to estimate mean diffusivity (MD), fractional anisotropy (FA), mean kurtosis (MK) and its components-the anisotropic and isotropic kurtoses (MKA and MKI). Meningioma consistency was estimated for 16 patients during resection based on ultrasonic aspiration intensity, ease of resection with instrumentation or suction. Grade and type were determined by histopathological analysis. The relation between consistency, grade and type and dMRI parameters was analyzed inside the tumor ("whole-tumor") and within brain tissue in the immediate periphery outside the tumor ("rim") by histogram analysis.RESULTS: Lower 10th percentiles of MK and MKA in the whole-tumor were associated with firm consistency compared with pooled soft and variable consistency (n = 7 vs 9; U test, p = 0.02 for MKA 10 and p = 0.04 for MK10) and lower 10th percentile of MD with variable against soft and firm (n = 5 vs 11; U test, p = 0.02). Higher standard deviation of MKI in the rim was associated with lower grade (n = 22 vs 8; U test, p = 0.04) and in the MKI maps we observed elevated rim-like structure that could be associated with grade. Higher median MKA and lower median MKI distinguished psammomatous type from other pooled meningioma types (n = 5 vs 25; U test; p = 0.03 for MKA 50 and p = 0.03 and p = 0.04 for MKI 50).CONCLUSION: Parameters from tensor-valued dMRI can facilitate prediction of consistency, grade and type.
|
|
