Book ChapterDOI
Ischemic Stroke Lesion Segmentation in CT Perfusion Images Using U-Net with Group Convolutions
Anish Monsley Kirupakaran,Bala Chakravarthy Neelapu,Rabul Hussain Laskar +2 more
- pp 276-288
TLDR
In this article , a deep learning model derived from U-Net was proposed to process all the perfusion parameter maps parallelly at the same time independently, which helps in avoiding the necessity of developing and training different models to process perfusion maps independently.Abstract:
Ischemic stroke is a cerebrovascular disease caused by a blockage in blood vessels of the brain. The early detection of the stroke helps in preventing the penumbra from turning into the core. So, early detection is essential. But the variability of the stroke lesion in size, location, and appearance makes the automatic segmentation of the stroke lesion difficult. Computed Tomography Perfusion (CTP) is more suitable because of its wide availability and the less acquisition time as compared to Magnetic Resonance Imaging (MRI). CTP parameter maps include Cerebral Blood Volume (CBV), Cerebral Blood Flow (CBF), Mean Transit Time (MTT), and Time to Peak (Tmax). In this paper, we propose a deep learning model derived from U-Net that can process all the perfusion parameter maps parallelly at the same time independently. This architecture helps in avoiding the necessity of developing and training different models to process the perfusion maps independently. The significant modifications in the proposed model are i) incorporation of group convolutions to process the parameter maps separately and ii) introduced element-wise summation of feature maps instead of concatenation. Also, the class imbalance problem in medical datasets makes the segmentations more challenging. This is overcome by employing a loss that is a combination of cross entropy and soft dice loss. The model is trained from scratch. We performed a 5-fold cross-validation on the data. The proposed model achieves the highest 0.441 as the dice coefficient in one fold and the average dice score is 0.421. The experimentation is conducted on Ischemic Stroke Lesion Segmentation Challenge (ISLES) 2018 dataset. read more
References
More filters
Book ChapterDOI
U-Net: Convolutional Networks for Biomedical Image Segmentation
TL;DR: Neber et al. as discussed by the authors proposed a network and training strategy that relies on the strong use of data augmentation to use the available annotated samples more efficiently, which can be trained end-to-end from very few images and outperforms the prior best method (a sliding-window convolutional network) on the ISBI challenge for segmentation of neuronal structures in electron microscopic stacks.
Journal ArticleDOI
Deconvolution-based CT and MR brain perfusion measurement: theoretical model revisited and practical implementation details
TL;DR: This paper presents a comprehensive derivation and explanation of the underlying physiological model for intravascular tracer systems, and discusses the need for regularization in order to obtain physiologically reasonable results.
Journal ArticleDOI
Global Epidemiology of Stroke and Access to Acute Ischemic Stroke Interventions.
TL;DR: In this paper, the authors provide an up-to-date review of the incidence of stroke and large vessel occlusion around the globe, as well as the eligibility and access to IV thrombolysis (IVT) and mechanical thrombectomy (MT) worldwide.
Book ChapterDOI
Dense Multi-path U-Net for Ischemic Stroke Lesion Segmentation in Multiple Image Modalities
TL;DR: In this paper, the authors focus on the problem of delineating infarcted tissue in ischemic stroke lesions to determine the extend of damage and optimal treatment for this life-threatening condition.
Journal ArticleDOI
Towards Clinical Diagnosis: Automated Stroke Lesion Segmentation on Multi-Spectral MR Image Using Convolutional Neural Network
TL;DR: Wang et al. as mentioned in this paper proposed a 2D-slice-based segmentation method using a residual-structured fully convolutional network (Res-FCN), which achieved a mean dice coefficient of 0.645 with a mean number of false negative lesions of 1.515 per subject.