Improvement of automatic ischemic stroke lesion segmentation in CT perfusion maps using a learned deep neural network

There is an error in the Conclusions section of the manuscript. The entire Conclusions section was not included. The Conclusions section should read: In this work, nine different classifiers were used for ischemic stroke lesion segmentation from brain MRI images and evaluated using different ground truth sets and scenarios. Based on the results of this study, it seems justified to recommend RDF classifiers as the basis for method development, as they are fast, stable, and robust. Within this context, alternative features, as well as better pre- and post-processing methods should be investigated. Ischemic stroke lesion segmentation is a difficult problem with uncertain ground truth and a strong dependency on the pre-processing methods. Hence, improvements in this area are as important as developing better classifiers and features. While the obtained RDF classification results outperform all previously published methods, human observer accuracy is not yet reached and ischemic stroke lesion segmentation remains a complicated problem. Apart from RDF classifiers, convolutional neural networks appear to hold more potential for improvement and should be employed in use-cases where accuracy is considered more important than speed, usability, and ease-of-configuration. While this study compared different classifier solutions for ischemic stroke lesion segmentation in detail, the results have been obtained under the premise of a single type of pre-processing and a fixed set of features (except in the case of the CNN). It would be desirable to investigate the influence of different features and pre-processing decisions on the segmentation results. Furthermore, the presented methods have been devised and implemented by a single team of researchers. An open-for-all comparison, as e.g. the scheduled ISLES Challenge at the forthcoming MICCAI 2015 conference, will provide a greater insight in the difficult problem of stroke lesion segmentation. There is an error in Table 2 as well as Table 3. The caption for each table is missing. The authors have provided the corrected version here. Table 2 Flair scenario Table 3 Besteffort scenario Trained with GTG, evaluated on GTG, average computed over 31/37 cases, stars denote significant difference to best-performing method (in bold) with ** = p < 0.01 and * = p < 0.05, train-times given for a single training round, value after ± denotes the standard deviation. Trained with GTG, evaluated on GTG, average computed over 33/37 cases, stars denote significant difference to best-performing method (in bold) with ** = p < 0.01 and * = p < 0.05, value after ± denotes the standard deviation.

Ischemic stroke lesion segmentation using stacked sparse autoencoder

Ischemic stroke segmentation at an acute stage is vital in assessing the severity of patients' impairment and guiding therapeutic decision-making for reperfusion. Although many deep learning studies have shown attractive performance in medical segmentation, it is difficult to use these models trained on public data with private hospitals' datasets. Here, we demonstrate an ensemble model that employs two different multimodal approaches for generalization, a more effective way to perform on external datasets. First, after we jointly train a segmentation model on diffusion-weighted imaging (DWI) and apparent diffusion coefficient (ADC) MR modalities, the model is inferred on the DWI images. Second, a channel-wise segmentation model is trained by concatenating the DWI and ADC images as input, and then is inferred using both MR modalities. Before training with ischemic stroke data, we utilized BraTS 2021, a public brain tumor dataset, for transfer learning. An extensive ablation study evaluates which strategy learns better representations for ischemic stroke segmentation. In our study, nnU-Net well-known for robustness is selected as our baseline model. Our proposed method is evaluated on three different datasets: the Asan Medical Center (AMC) I and II, and the 2022 Ischemic Stroke Lesion Segmentation (ISLES). Our experiments are widely validated over a large, multi-center, and multi-scanner dataset with a huge amount of 846 scans. Not only stroke lesion models can benefit from transfer learning using brain tumor data, but combining the MR modalities using different training schemes also highly improves segmentation performance. The method achieved a top-1 ranking in the ongoing ISLES'22 challenge and performed particularly well on lesion-wise metrics of interest to neuroradiologists, achieving a Dice coefficient of 78.69% and a lesion-wise F1 score of 82.46%. Also, the method was relatively robust on the AMC I (Dice, 60.35%; lesion-wise F1, 68.30%) and II (Dice; 74.12%; lesion-wise F1, 67.53%) datasets in different settings. The high segmentation accuracy of our proposed method could improve radiologists' ability to detect ischemic stroke lesions in MRI images. Our model weights and inference code are available on https://github.com/MDOpx/ISLES22-model-inference .

Although deep learning methods have been widely applied in medical image lesion segmentation, it is still challenging to apply it for segmenting ischemic stroke lesions, which are different from brain tumors in lesion characteristics, segmentation difficulty, algorithm maturity, and segmentation accuracy. Three main stages are used to describe the manifestations of stroke. For acute ischemic stroke, the size of the lesions is similar to that of the brain tumor, and the current deep learning methods have been able to achieve a high segmentation accuracy. For sub-acute and chronic ischemic stroke, the segmentation results of mainstream deep learning algorithms are still unsatisfactory as lesions in these stages are small and diffuse. By using three scientific search engines including CNKI, Web of Science and Google Scholar, this paper aims to comprehensively understand the state-of-the-art deep learning algorithms applied to segment ischemic stroke lesions. For the first time, this paper discusses the current situation, challenges, and development directions of deep learning algorithms applied to ischemic stroke lesion segmentation in different stages. In the future, a system that can directly identify different stroke stages and automatically select the suitable network architecture for the stroke lesion segmentation needs to be proposed.

Infarcted brain tissue resulting from acute stroke readily shows up as hyperintense regions within diffusion-weighted magnetic resonance imaging (DWI). It has also been proposed that computed tomography perfusion (CTP) could alternatively be used to triage stroke patients, given improvements in speed and availability, as well as reduced cost. However, CTP has a lower signal to noise ratio compared to MR. In this work, we investigate whether a conditional mapping can be learned by a generative adversarial network to map CTP inputs to generated MR DWI that more clearly delineates hyperintense regions due to ischemic stroke. We detail the architectures of the generator and discriminator and describe the training process used to perform image-to-image translation from multi-modal CT perfusion maps to diffusion weighted MR outputs. We evaluate the results both qualitatively by visual comparison of generated MR to ground truth, as well as quantitatively by training fully convolutional neural networks that make use of generated MR data inputs to perform ischemic stroke lesion segmentation. Segmentation networks trained using generated CT-to-MR inputs result in at least some improvement on all metrics used for evaluation, compared with networks that only use CT perfusion input.

MR images (MRIs) accurate segmentation of brain lesions is important for improving cancer diagnosis, surgical planning, and prediction of outcome. However, manual and accurate segmentation of brain lesions from 3D MRIs is highly expensive, time-consuming, and prone to user biases. We present an efficient yet conceptually simple brain segmentation network (referred as Brain SegNet), which is a 3D residual framework for automatic voxel-wise segmentation of brain lesion. Our model is able to directly predict dense voxel segmentation of brain tumor or ischemic stroke regions in 3D brain MRIs. The proposed 3D segmentation network can run at about 0.5s per MRIs - about 50 times faster than previous approaches Med Image Anal 43: 98–111, 2018, Med Image Anal 36:61–78, 2017. Our model is evaluated on the BRATS 2015 benchmark for brain tumor segmentation, where it obtains state-of-the-art results, by surpassing recently published results reported in Med Image Anal 43: 98–111, 2018, Med Image Anal 36:61–78, 2017. We further applied the proposed Brain SegNet for ischemic stroke lesion outcome prediction, with impressive results achieved on the Ischemic Stroke Lesion Segmentation (ISLES) 2017 database.

A lightweight asymmetric U-Net framework for acute ischemic stroke lesion segmentation in CT and CTP images

Diffusion-weighted magnetic resonance imaging (DWI) is sensitive to acute ischemic stroke and is a common diagnostic method for the stroke. However, the diagnostic result relies on the visual observation of neurologists which may vary from doctor to doctor under different circumstance. And manual segmentation is often a time-consuming and subjective process. The time from onset to thrombus removal has a significant impact on the prognosis of patients with acute ischemic stroke. The shorter the time, the better the prognosis. For this purpose we present a novel framework to quickly and automatically segment the ischemic stroke lesions on DWI. We mainly have three contributions: firstly, we design a detection and segmentation network (DSN) to solve the two kinds of data imbalance; secondly, we propose a triple-branch DSN architecture, used for extracting different plane feature respectively; thirdly, we propose a multi-plane fusion network (MPFN), which aims to make final prediction more accurate. Extensive experiments on ISLES2015 SSIS DWI sequence dataset demonstrate the superiority of our proposed segmentation method. The dice reached 62.2% and the sensitivity reached 71.7%.

Brain stroke is the major second leading cause of death for the people above the age of 60 and fifth leading cause in people aged 15–59. Automatic ischemic stroke lesion segmentation of Magnetic Resonance Images (MRI) is an important task since manual identification by medical experts is time-consuming. In this work, the Diffusion Weighted (DW) brain MR images of ISLES 2015 challenge dataset and Radiopaedia are pre-processed based on hybrid contrast enhancement and filtering technique. The skull is stripped by thresholding techniques along with mathematical morphology procedure. Later the identification of ischemic stroke lesion is performed by maximum entropy-based and Fuzzy C Means (FCM) thresholding. The segmented results are compared with expert’s ground truth and the performance is analyzed by means of similarity indices. It is observed that FCM performs better in identifying multiple and minute stroke lesion in MR images. There is nearly 98% close match between FCM segmented results and the ground truth scribed by the experts. Thus the proposed automated work flow helps in assisting the clinicians to identify the ischemic stroke lesion in DW modality of brain MR images.

A life threatening medical condition occurs when arteries that supplies blood to the brain gets blocked resulting in Ischemic Stroke. Magnetic resonance imaging (MRI) plays major role in diagnosis of brain stroke at early stages. Manual detection of stroke lesions by medical experts is time-consuming, expensive, and susceptible to intra- and inter-observer variability. Accurate detection of stroke lesions from brain MRI, the challenging task requires development of automated computer aided diagnostic techniques. This paper aims at reviewing the state of art techniques currently available fulfilling the above objectives, their merits and limitation. Through this review we figure out the modifications that need to be carried out in future to develop best automated diagnostic tool which performs better and mitigates all the pitfalls in current literatures.

CT perfusion imaging is important in the imaging workup of acute ischemic stroke for evaluating affected cerebral tissue. CT perfusion analysis software produces cerebral perfusion maps from commonly noisy spatio-temporal CT perfusion data. High levels of noise can influence the results of CT perfusion analysis, necessitating software tuning. This work proposes a novel approach for CT perfusion analysis that uses physics-informed learning, an optimization framework that is robust to noise. In particular, we propose SPPINN: Spatio-temporal Perfusion Physics-Informed Neural Network and research spatio-temporal physics-informed learning. SPPINN learns implicit neural representations of contrast attenuation in CT perfusion scans using the spatio-temporal coordinates of the data and employs these representations to estimate a continuous representation of the cerebral perfusion parameters. We validate the approach on simulated data to quantify perfusion parameter estimation performance. Furthermore, we apply the method to in-house patient data and the public Ischemic Stroke Lesion Segmentation 2018 benchmark data to assess the correspondence between the perfusion maps and reference standard infarct core segmentations. Our method achieves accurate perfusion parameter estimates even with high noise levels and differentiates healthy tissue from infarcted tissue. Moreover, SPPINN perfusion maps accurately correspond with reference standard infarct core segmentations. Hence, we show that using spatio-temporal physics-informed learning for cerebral perfusion estimation is accurate, even in noisy CT perfusion data. The code for this work is available at https://github.com/lucasdevries/SPPINN.

For patients with acute ischemic stroke (AIS), rapid and accurate lesion localization is critical for improving treatment outcomes. However, automatic stroke lesion segmentation remains highly challenging due to the scarcity of large-scale annotated datasets. Recently, semi-supervised learning (SSL) has achieved remarkable progress in medical image segmentation, yet its performance is still hindered by unreliable pseudo-labels. To address this issue, we propose a novel SSL framework, termed reliability-aware mutual learning (RAML), which employs two subnetworks with a shared encoder, a primary decoder, and an auxiliary decoder. Specifically, RAML introduces uncertain region relearning (URR) regularization, which leverages prediction uncertainty from both subnetworks to identify and refine unreliable regions in labeled images. For unlabeled images, reliability-aware mutual pseudo-supervision (RMPS) regularization is designed to enable cross-supervision based on reliable pseudo-labels. Furthermore, feature difference learning (FDL) regularization is incorporated to promote prediction diversity across subnetworks. Experiments on two acute ischemic stroke datasets and the Left Atrium dataset demonstrate the effectiveness of the proposed RAML in semi-supervised segmentation tasks. The code for this project is available at https://github.com/EricMedimuist/RAML.

Early detection of newly emerging diseases, lesion severity assessment, differentiation of medical conditions and automated screening are examples for the wide applicability and importance of anomaly detection (AD) and unsupervised segmentation in medicine. Normal fine-grained tissue variability such as present in pulmonary anatomy is a major challenge for existing generative AD methods. Here, we propose a novel generative AD approach addressing this issue. It consists of an image-to-image translation for anomaly-free reconstruction and a subsequent patch similarity scoring between observed and generated image-pairs for precise anomaly localization. We validate the new method on chest computed tomography (CT) scans for the detection and segmentation of infectious disease lesions. To assess generalizability, we evaluate the method on an ischemic stroke lesion segmentation task in T1-weighted brain MRI. Results show improved pixel-level anomaly segmentation in both chest CTs and brain MRIs, with relative DICE score improvements of +1.9% and +4.4%, respectively, compared to other state-of-the-art reconstruction-based methods.

Analysis of multimodal fusion strategies in deep learning for ischemic stroke lesion segmentation on computed tomography perfusion data

Background CT can provide information regarding myocardial perfusion and expansion of the extracellular space, which is relevant to patients with cardiac amyloidosis (CA). Purpose To evaluate the role of CT in the diagnosis and prognosis of CA. Materials and Methods In this prospective study (Commission National de l'Informatique et des Libertés registration no. 1431858), participants with CA, participants with nonamyloid cardiac hypertrophy (NACH), and participants without hypertrophy were included between April 2017 and December 2018. The confirmed diagnosis of CA was determined according to established criteria (ie, proven with positive bone scintigraphy or endomyocardial biopsy). All participants were imaged with dynamic CT perfusion imaging at whole-heart cardiac CT. Extracellular volume measured at CT and myocardial perfusion parameters calculated on CT perfusion maps were compared among different participant groups. Differences between continuous data were tested using the unpaired t test, Mann-Whitney rank-sum test, or the Kruskal-Wallis test. Results A total of 84 participants with CA, 43 participants with NACH, and 33 participants without hypertrophy were included. Participants with CA exhibited a higher value of extracellular volume measured at CT (mean, 54.7% ± 9.7 [standard deviation]) than participants with NACH (mean, 34.6% ± 9.1; P < .001) and participants without hypertrophy (mean, 35.9% ± 9.9; P = .001). Mean myocardial blood volume and mean myocardial blood flow were lower in participants with CA (mean myocardial blood volume: 4.05 mL/100 g of myocardium ± 0.80; mean myocardial blood flow: 73.2 mL/100 g of myocardium per minute ± 25.7) compared to participants with NACH (mean myocardial blood volume: 5.38 mL/100 g of myocardium ± 1.20, P < .001; mean myocardial blood flow: 89.6 mL/100 g of myocardium per minute ± 31.3, P = .007) and participants without hypertrophy (mean myocardial blood volume: 5.68 mL/100 g of myocardium ± 1.05; mean myocardial blood flow: 106.3 mL/100 g of myocardium per minute ± 29.8; P < .001 for both). Extracellular volume measured at CT (hazard ratio >0.56 vs ≤0.56 = 4.2 [95% CI: 1.4, 11.8]), mean slope (hazard ratio ≤3.0 sec-1 vs >3.0 sec-1 = 0.2 [95% CI: 0.1, 0.8]), and time to peak (hazard ratio >20 seconds vs ≤20 seconds = 11.6 [95% CI: 1.3, 101.6]) were predictive of mortality in participants with CA. Conclusion Participants with cardiac amyloidosis exhibited an increase in extracellular volume at CT and abnormal CT perfusion parameters. Extracellular volume and several perfusion parameters were predictive of mortality. © RSNA, 2021 Online supplemental material is available for this article. See also the editorial by Zimmerman in this issue.