Automatic prediction of infarct growth in acute ischemic stroke from MR apparent diffusion coefficient maps

Predicting Hypoperfusion Lesion and Target Mismatch in Stroke from Diffusion-weighted MRI Using Deep Learning

Background Perfusion imaging is important to identify a target mismatch in stroke but requires contrast agents and postprocessing software. Purpose To use a deep learning model to predict the hypoperfusion lesion in stroke and identify patients with a target mismatch profile from diffusion-weighted imaging (DWI) and clinical information alone, using perfusion MRI as the reference standard. Materials and Methods Imaging data sets of patients with acute ischemic stroke with baseline perfusion MRI and DWI were retrospectively reviewed from multicenter data available from 2008 to 2019 (Imaging Collaterals in Acute Stroke, Diffusion and Perfusion Imaging Evaluation for Understanding Stroke Evolution 2, and University of California, Los Angeles stroke registry). For perfusion MRI, rapid processing of perfusion and diffusion software automatically segmented the hypoperfusion lesion (time to maximum, ≥6 seconds) and ischemic core (apparent diffusion coefficient [ADC], ≤620 × 10-6 mm2/sec). A three-dimensional U-Net deep learning model was trained using baseline DWI, ADC, National Institutes of Health Stroke Scale score, and stroke symptom sidedness as inputs, with the union of hypoperfusion and ischemic core segmentation serving as the ground truth. Model performance was evaluated using the Dice score coefficient (DSC). Target mismatch classification based on the model was compared with that of the clinical-DWI mismatch approach defined by the DAWN trial by using the McNemar test. Results Overall, 413 patients (mean age, 67 years ± 15 [SD]; 207 men) were included for model development and primary analysis using fivefold cross-validation (247, 83, and 83 patients in the training, validation, and test sets, respectively, for each fold). The model predicted the hypoperfusion lesion with a median DSC of 0.61 (IQR, 0.45-0.71). The model identified patients with target mismatch with a sensitivity of 90% (254 of 283; 95% CI: 86, 93) and specificity of 77% (100 of 130; 95% CI: 69, 83) compared with the clinical-DWI mismatch sensitivity of 50% (140 of 281; 95% CI: 44, 56) and specificity of 89% (116 of 130; 95% CI: 83, 94) (P < .001 for all). Conclusion A three-dimensional U-Net deep learning model predicted the hypoperfusion lesion from diffusion-weighted imaging (DWI) and clinical information and identified patients with a target mismatch profile with higher sensitivity than the clinical-DWI mismatch approach. ClinicalTrials.gov registration nos. NCT02225730, NCT01349946, NCT02586415 © RSNA, 2022 Supplemental material is available for this article. See also the editorial by Kallmes and Rabinstein in this issue.

Predicting final ischemic stroke lesions from initial diffusion-weighted images using a deep neural network

BACKGROUND

For prognosis of stroke, measurement of the diffusion-perfusion mismatch is a common practice for estimating tissue at risk of infarction in the absence of timely reperfusion. However, perfusion-weighted imaging (PWI) adds time and expense to the acute stroke imaging workup. We explored whether a deep convolutional neural network (DCNN) model trained with diffusion-weighted imaging obtained at admission could predict final infarct volume and location in acute stroke patients.

METHODS

In 445 patients, we trained and validated an attention-gated (AG) DCNN to predict final infarcts as delineated on follow-up studies obtained 3 to 7 days after stroke. The input channels consisted of MR diffusion-weighted imaging (DWI), apparent diffusion coefficients (ADC) maps, and thresholded ADC maps with values less than 620 × 10-6 mm2/s, while the output was a voxel-by-voxel probability map of tissue infarction. We evaluated performance of the model using the area under the receiver-operator characteristic curve (AUC), the Dice similarity coefficient (DSC), absolute lesion volume error, and the concordance correlation coefficient (ρc) of the predicted and true infarct volumes.

RESULTS

The model obtained a median AUC of 0.91 (IQR: 0.84-0.96). After thresholding at an infarction probability of 0.5, the median sensitivity and specificity were 0.60 (IQR: 0.16-0.84) and 0.97 (IQR: 0.93-0.99), respectively, while the median DSC and absolute volume error were 0.50 (IQR: 0.17-0.66) and 27 ml (IQR: 7-60 ml), respectively. The model's predicted lesion volumes showed high correlation with ground truth volumes (ρc = 0.73, p < 0.01).

CONCLUSION

An AG-DCNN using diffusion information alone upon admission was able to predict infarct volumes at 3-7 days after stroke onset with comparable accuracy to models that consider both DWI and PWI. This may enable treatment decisions to be made with shorter stroke imaging protocols.

Medical image analysis methods in MR/CT-imaged acute-subacute ischemic stroke lesion: Segmentation, prediction and insights into dynamic evolution simulation models. A critical appraisal

Over the last 15 years, basic thresholding techniques in combination with standard statistical correlation-based data analysis tools have been widely used to investigate different aspects of evolution of acute or subacute to late stage ischemic stroke in both human and animal data. Yet, a wave of biology-dependent and imaging-dependent issues is still untackled pointing towards the key question: "how does an ischemic stroke evolve?" Paving the way for potential answers to this question, both magnetic resonance (MRI) and CT (computed tomography) images have been used to visualize the lesion extent, either with or without spatial distinction between dead and salvageable tissue. Combining diffusion and perfusion imaging modalities may provide the possibility of predicting further tissue recovery or eventual necrosis. Going beyond these basic thresholding techniques, in this critical appraisal, we explore different semi-automatic or fully automatic 2D/3D medical image analysis methods and mathematical models applied to human, animal (rats/rodents) and/or synthetic ischemic stroke to tackle one of the following three problems: (1) segmentation of infarcted and/or salvageable (also called penumbral) tissue, (2) prediction of final ischemic tissue fate (death or recovery) and (3) dynamic simulation of the lesion core and/or penumbra evolution. To highlight the key features in the reviewed segmentation and prediction methods, we propose a common categorization pattern. We also emphasize some key aspects of the methods such as the imaging modalities required to build and test the presented approach, the number of patients/animals or synthetic samples, the use of external user interaction and the methods of assessment (clinical or imaging-based). Furthermore, we investigate how any key difficulties, posed by the evolution of stroke such as swelling or reperfusion, were detected (or not) by each method. In the absence of any imaging-based macroscopic dynamic model applied to ischemic stroke, we have insights into relevant microscopic dynamic models simulating the evolution of brain ischemia in the hope to further promising and challenging 4D imaging-based dynamic models. By depicting the major pitfalls and the advanced aspects of the different reviewed methods, we present an overall critique of their performances and concluded our discussion by suggesting some recommendations for future research work focusing on one or more of the three addressed problems.

Prediction of subacute infarct size in acute middle cerebral artery stroke: comparison of perfusion-weighted imaging and apparent diffusion coefficient maps

PURPOSE

To compare perfusion-weighted (PW) imaging and apparent diffusion coefficient (ADC) maps in prediction of infarct size and growth in patients with acute middle cerebral artery infarct.

MATERIALS AND METHODS

This study was approved by the local institutional review board. Written informed consent was obtained from all 80 patients. Subsequent infarct volume and growth on follow-up magnetic resonance (MR) images obtained within 6 days were compared with the predictions based on PW images by using a time-to-peak threshold greater than 4 seconds and ADC maps obtained less than 12 hours after middle cerebral artery infarct. ADC- and PW imaging-predicted infarct growth areas and infarct volumes were correlated with subsequent infarct growth and follow-up diffusion-weighted (DW) imaging volumes. The impact of MR imaging time delay on the correlation coefficient between the predicted and subsequent infarct volumes and individual predictions of infarct growth by using receiver operating characteristic curves were assessed.

RESULTS

The infarct volume measurements were highly reproducible (concordance correlation coefficient [CCC] of 0.965 and 95% confidence interval [CI]: 0.949, 0.976 for acute DW imaging; CCC of 0.995 and 95% CI: 0.993, 0.997 for subacute DW imaging). The subsequent infarct volume correlated (P<.0001) with ADC- (ρ=0.853) and PW imaging- (ρ=0.669) predicted volumes. The correlation was higher for ADC-predicted volume than for PW imaging-predicted volume (P<.005), but not when the analysis was restricted to patients without recanalization (P=.07). The infarct growth correlated (P<.0001) with PW imaging-DW imaging mismatch (ρ=0.470) and ADC-DW imaging mismatch (ρ=0.438), without significant differences between both methods (P=.71). The correlations were similar among time delays with ADC-predicted volumes but decreased with PW imaging-based volumes beyond the therapeutic window. Accuracies of ADC- and PW imaging-based predictions of infarct growth in an individual prediction were similar (area under the receiver operating characteristic curve [AUC] of 0.698 and 95% CI: 0.585, 0.796 vs AUC of 0.749 and 95% CI: 0.640, 0.839; P=.48).

CONCLUSION

The ADC-based method was as accurate as the PW imaging-based method for evaluating infarct growth and size in the subacute phase.

Magnetic resonance diffusion-perfusion mismatch in acute ischemic stroke: An update

The concept of magnetic resonance perfusion-diffusion mismatch (PDM) provides a practical and approximate measure of the tissue at risk and has been increasingly applied for the evaluation of hyperacute and acute stroke in animals and patients. Recent studies demonstrated that PDM does not optimally define the ischemic penumbra; because early abnormality on diffusion-weighted imaging overestimates the infarct core by including part of the penumbra, and the abnormality on perfusion weighted imaging overestimates the penumbra by including regions of benign oligemia. To overcome these limitations, many efforts have been made to optimize conventional PDM. Various alternatives beyond the PDM concept are under investigation in order to better define the penumbra. The PDM theory has been applied in ischemic stroke for at least three purposes: to be used as a practical selection tool for stroke treatment; to test the hypothesis that patients with PDM pattern will benefit from treatment, while those without mismatch pattern will not; to be a surrogate measure for stroke outcome. The main patterns of PDM and its relation with clinical outcomes were also briefly reviewed. The conclusion was that patients with PDM documented more reperfusion, reduced infarct growth and better clinical outcomes compared to patients without PDM, but it was not yet clear that thrombolytic therapy is beneficial when patients were selected on PDM. Studies based on a larger cohort are currently under investigation to further validate the PDM hypothesis.

Hyperglycemia and the fate of apparent diffusion coefficient-defined ischemic penumbra

BACKGROUND AND PURPOSE

Previous data have shown the feasibility of identifying ischemic penumbra in patients with acute stroke by using a semiautomated analysis of ADC maps. Here, we investigated whether the fate of ADC-defined penumbra was altered by HG. We also examined the interaction between HG and arterial recanalization on infarct growth.

MATERIALS AND METHODS

We examined 94 patients by using MR imaging within 6 hours of stroke onset and a follow-up MR imaging within 7 days. The ADC-defined tissue-at-risk was calculated from the early MR imaging. Patients were classified according to high (>7 mmol/L; n = 34/94, HG) or normal (n = 60/94) baseline SGL. The impact of HG status on infarct growth was assessed by using multiple regression models and analysis of the slopes of regression lines for each group. Interaction between HG status and arterial recanalization on infarct growth was investigated by using multiple regression analysis.

RESULTS

The slope of the predicted versus observed infarct growth regression line was steeper in HG than non-HG patients (P = .0008), suggesting that infarct growth within ADC-defined tissue-at-risk was increased in HG patients. The effect was 2.8 times more severe in nonrecanalized patients (P = .01) than in patients with recanalization (P = .001).

CONCLUSIONS

ADC-defined tissue-at-risk may represent ischemic penumbra because part of this area may be salvaged in normal SGL patients. The toxicity in HG patients seems to be more related to penumbra-infarction transition than reperfusion injury in humans because the effect was larger in nonrecanalized than in recanalized patients.

Prediction of infarct core and salvageable ischemic tissue volumes by analyzing apparent diffusion coefficient without intravenous contrast material

RATIONALE AND OBJECTIVES

To investigate whether baseline apparent diffusion coefficient (ADC) maps can be employed to predict both infarct core and salvageable ischemic tissue volumes in acute ischemic stroke.

MATERIALS AND METHODS

An automated image analysis system based on baseline ADC maps was tested against 30 patients with acute ischemic stroke of anterior circulation to predict both infarct core and salvageable ischemic tissue volumes. The predicted infarct core and predicted salvageable ischemic tissue were quantitatively and qualitatively compared with follow-up imaging data in recanalization and no recanalization groups, respectively. Direct comparisons with perfusion- and diffusion- weighted magnetic resonance imaging measures were also made. Wilcoxon signed-rank test, Spearman rank correlation, and Bland-Altman plots were performed.

RESULTS

In the recanalization group, the predicted infarct core volume was significantly correlated with the final infarct volume (r = 0. 868, P < .001). In the no recanalization group, the predicted final infarct volume (sum of the predicted infarct core and salvageable ischemic tissue volumes), as well as the predicted salvageable ischemic tissue volume, was also significantly correlated with the true final infarct volume (r = 0.955, P < .001) and infarct growth (r = 0.918, P < .001), respectively. The volumes of perfusion-diffusion mismatch were significantly larger than those of infarct growth and predicted salvageable ischemic tissue. Good agreement between predicted and true final infarct lesions was visualized by Bland-Altman plots in two groups. Direct visual comparative analysis revealed good qualitative agreement between the true final infarct and predicted lesions in 21 patients.

CONCLUSION

The proposed ADC based approach may be a feasible and practical tool to predict the volumes of infarct core and salvageable ischemic tissue without intravenous contrast media-enhanced perfusion-weighted imaging at baseline.

Whole brain perfused blood volume CT: visualization of infarcted tissue compared to quantitative perfusion CT

RATIONALE AND OBJECTIVES

This study determines the value of whole brain color-coded three-dimensional perfused blood volume (PBV) computed tomography (CT) for the visualization of the infarcted tissue in acute stroke patients.

MATERIALS AND METHODS

Nonenhanced CT (NECT), perfusion CT (PCT), and CT angiography (CTA) in 48 patients with acute ischemic stroke were performed. Whole brain PBV was calculated from NECT and CTA data sets using commercial software. PBV slices in identical orientation to the PCT slices were reconstructed and the area of visual perfusion abnormality on PBV maps was measured. The infarct core in the corresponding PCT slices (CBV <2.0 mL/100 g) was measured automatically with commercial software. The ischemic area on PBV and the infarct core on quantitative PCT were compared using the Pearsons-R correlation coefficient. Significance was considered for P < .05.

RESULTS

The quantitative PCT demonstrated a mean infarct core volume of 35.48 +/- 32.17 cm(3), whereas the volume of visual perfusion abnormality of the corresponding PBV slices was 37.16 +/- 37.59 cm(3). The perfusion abnormality in PBV was highly correlated with the infarct core of quantitative PCT for area per slice (r = 0.933, P < .01) as well as volume (r = 0.922, P < .01).

CONCLUSIONS

PBV can serve as surrogate marker corresponding to the infarct core in acute stroke with whole brain coverage.

[Mechanical recanalization]

Although several studies and registries have focused on new interventional systems for the treatment of acute ischemic stroke, a standard procedure has not yet been established. The procedure itself is still controversially discussed but studies have shown that patients who were successfully treated with mechanical recanalization had a better clinical outcome.

Robust nonparametric segmentation of infarct lesion from diffusion-weighted MR images

Magnetic Resonance Imaging (MRI) is increasingly used for the diagnosis and monitoring of neurological disorders. In particular Diffusion-Weighted MRI (DWI) is highly sensitive in detecting early cerebral ischemic changes in acute stroke. Cerebral infarction lesion segmentation from DWI is accomplished in this work by applying nonparametric density estimation. The quality of the class boundaries is improved by including an edge confidence map, that is the confidence of truly being in the presence of a border between adjacent regions. The adjacency graph, that is constructed with the label regions, is analyzed and pruned to merge adjacent regions. The method was applied to real images, keeping all parameters constant throughout the process for each data set. The combination of region segmentation and edge detection proved to be a robust automatic technique of segmentation from DWI images of cerebral infarction regions in acute ischemic stroke. In a comparison with the reference infarct lesions segmentation, the automatic segmentation presented a significant correlation (r=0.935), and an average Tanimoto index of 0.538.