124 research outputs found
Functional Outcome Prediction in Acute Ischemic Stroke
Get PDFBACKGROUND Stroke is one of the most prevalent neurological diseases and causes of disability worldwide. Functional outcome prediction models can assist the treatment decision process and optimize acute ischemic stroke health care. Current models often use a limited set of input features to predict functional outcome, although combining various types of features could improve model performance. Furthermore, they often incorporate follow-up information, while prediction models applicable in the acute setting are desirable. METHODS We trained an ensemble model consisting of five machine learning models with leave-one-out cross-validation to predict the binarized modified Rankin Scale score three months after stroke onset in patients with acute ischemic stroke caused by a large vessel occlusion who received endovascular treatment. We used clinical variables, treatment variables and lesion loads derived from registration of a stroke population-specific neuroanatomical CT brain atlas with the follow-up non-contrast enhanced CT scan as input features. RESULTS Taking into account five performance metrics (accuracy, AUC, sensitivity, specificity and F1-score), the ensemble model and support vector machine (SVM) seemed to achieve the best performances out of the six models (ensemble model and the five individual machine learning models), with AUC values up to 0.76 and 0.77 respectively. The highest accuracy obtained with the ensemble model was 0.69, and with the SVM 0.72. Little variance in performance was found between the various sets of input features. CONCLUSION Although similar performances compared to current literature were obtained, conventional machine learning models might not be sophisticated enough to capture the complex interactions between input features for functional outcome prediction in acute ischemic stroke
Deep Learning for Ischemic Penumbra Segmentation from MR Perfusion Maps: Robustness to the Deconvolution Algorithm
No full textDetermining the penumbra, i.e., the at-risk but salvageable tissue, is crucial in the context of acute ischemic stroke imaging. Deep learning methods performing segmentation from perfusion parameter maps have shown promise in this regard. However, these methods rely on the computation of parameter maps via deconvolution algorithms, raising concerns about their generalizability across different medical centers. This study investigates the robustness of segmentation methods given different perfusion processing algorithms for dynamic susceptibility contrast magnetic resonance perfusion imaging. A neural network is first trained on a dataset of 94 patients with paired Tmax maps from a single MR perfusion algorithm, together with manual perfusion deficit segmentations. The network’s outputs are then compared on a second dataset of 268 patients, where Tmax inputs are generated with three different deconvolution algorithms. DICE coefficient along with the difference between estimated perfusion deficit volumes are used to quantify the agreement between predictions. Our findings demonstrate high variability in the predicted penumbra, even when Tmax inputs exhibit high similarity (SSIM > 0.8). This study therefore highlights the importance of exploring deconvolution-free methods to address the robustness issue for learning-based penumbra segmentation.LTS
Multimodal Deep Learning for Functional Outcome Prediction in Endovascular Therapy
Get PDFThe efficacy of endovascular therapy (EVT) in large vessel occlusion (LVO) of the anterior circulation depends on adequate patient selection. Patients can be selected based on their predicted functional outcome after EVT. Using a dataset composed of 1929 patients, we compare the functional outcome prediction performance of clinical baseline models, including the clinically validated MR PREDICTS decision tool, with an imaging based pipeline and a multimodal approach. The predicted outcome measure is dichotomized modified Rankin Scale score 90 days after mechanical thrombectomy. Binary classifier performance is quantified using Area-Under the receiver operating characteristic Curve (AUC). Combining clinical features with information extracted from CTA images does not significantly improve the performance of functional outcome prediction methods compared to the baseline model. This multimodal approach can however replace radiologically derived biomarkers, as its performance is non-inferior
The Detection and Segmentation of Blush in the Lenticulostriate Territory
Get PDFThe lenticulostriate territory is a region in the brain that is only supplied by lenticulostriate vessels. As this region does not benefit from collateral blood flow, it warrants attention in the event of ischemic stroke. Perfusion in the lenticulostriate territory shows up as a blush in digital subtraction angiography. Although this blush is observable, visual inspection is subjective and qualitative while a quantitative figure of merit is desired. To allow quantitative analysis of this blush, a segmentation of the correct blush is necessary. In this paper, a deep-learning approach is proposed to perform the novel segmentation of the blush in the lenticulostriate territory. To provide a first quantification of the blush, the hemisphere in which it occurred is also segmented. The ratio between these segmentations is a quantification of the size of the blush compared to the size of the hemisphere. Results indicate proof of concept, but more steps are needed before clinical application
Deep Quality Estimation: Creating Surrogate Models for Human Quality Ratings
Get PDFHuman ratings are abstract representations of segmentation quality. To
approximate human quality ratings on scarce expert data, we train surrogate
quality estimation models. We evaluate on a complex multi-class segmentation
problem, specifically glioma segmentation, following the BraTS annotation
protocol. The training data features quality ratings from 15 expert
neuroradiologists on a scale ranging from 1 to 6 stars for various
computer-generated and manual 3D annotations. Even though the networks operate
on 2D images and with scarce training data, we can approximate segmentation
quality within a margin of error comparable to human intra-rater reliability.
Segmentation quality prediction has broad applications. While an understanding
of segmentation quality is imperative for successful clinical translation of
automatic segmentation quality algorithms, it can play an essential role in
training new segmentation models. Due to the split-second inference times, it
can be directly applied within a loss function or as a fully-automatic dataset
curation mechanism in a federated learning setting.Comment: 10 pages, 5 figure
An approach to automatic object tracking system by combination of SIFT and RANSAC with mean shift and KLT
No full textBrain Tumor Segmentation Based on Non Negative Matrix Factorization and Fuzzy Clustering
No full text
- …
