Investigation of convolutional neural networks using multiple computed tomography perfusion maps to identify infarct core in acute ischemic stroke patients

Evaluating the effect of noise reduction strategies in CT perfusion imaging for predicting infarct core with deep learning

This study evaluates the efficacy of deep learning models in identifying infarct tissue on computed tomography perfusion (CTP) scans from patients with acute ischemic stroke due to large vessel occlusion, specifically addressing the potential influence of varying noise reduction techniques implemented by different vendors. We analyzed CTP scans from 60 patients who underwent mechanical thrombectomy achieving a modified thrombolysis in cerebral infarction (mTICI) score of 2c or 3, ensuring minimal changes in the infarct core between the initial CTP and follow-up MR imaging. Noise reduction techniques, including principal component analysis (PCA), wavelet, non-local means (NLM), and a no denoising approach, were employed to create hemodynamic parameter maps. Infarct regions identified on follow-up diffusion-weighted imaging (DWI) within 48 hours were co-registered with initial CTP scans and refined with ADC maps to serve as ground truth for training a data-augmented U-Net model. The performance of this convolutional neural network (CNN) was assessed using Dice coefficients across different denoising methods and infarct sizes, visualized through box plots for each parameter map. Our findings show no significant differences in model accuracy between PCA and other denoising methods, with minimal variation in Dice scores across techniques. This study confirms that CNNs are adaptable and capable of handling diverse processing schemas, indicating their potential to streamline diagnostic processes and effectively manage CTP input data quality variations.

Reducing False-Positives in CT Perfusion Infarct Core Segmentation Using Contralateral Local Normalization

BACKGROUND AND PURPOSE

The established global threshold of rCBF <30% for infarct core segmentation can lead to false-positives, as it does not account for the differences in blood flow between GM and WM and patient-individual factors, such as microangiopathy. To mitigate this problem, we suggest normalizing each voxel not only with a global reference value (ie, the median value of normally perfused tissue) but also with its local contralateral counterpart.

MATERIALS AND METHODS

We retrospectively enrolled 2830 CTP scans with suspected ischemic stroke, of which 335 showed obvious signs of microangiopathy. In addition to the conventional, global normalization, a local normalization was performed by dividing the rCBF maps with their mirrored and smoothed counterpart, which sets each voxel value in relation to the contralateral counterpart, intrinsically accounting for GM and WM differences and symmetric patient individual microangiopathy. Maps were visually assessed and core volumes were calculated for both methods.

RESULTS

Cases with obvious microangiopathy showed a strong reduction in false-positives by using local normalization (mean 14.7 mL versus mean 3.7 mL in cases with and without microangiopathy). On average, core volumes were slightly smaller, indicating an improved segmentation that was more robust against naturally low blood flow values in the deep WM.

CONCLUSIONS

The proposed method of local normalization can reduce overestimation of the infarct core, especially in the deep WM and in cases with obvious microangiopathy. False-positives in CTP infarct core segmentation might lead to less-than-optimal therapy decisions when not correctly interpreted. The proposed method might help mitigate this problem.

Artificial Intelligence and Machine Learning in the Diagnosis and Management of Stroke: A Narrative Review of United States Food and Drug Administration-Approved Technologies

Stroke is an emergency in which delays in treatment can lead to significant loss of neurological function and be fatal. Technologies that increase the speed and accuracy of stroke diagnosis or assist in post-stroke rehabilitation can improve patient outcomes. No resource exists that comprehensively assesses artificial intelligence/machine learning (AI/ML)-enabled technologies indicated for the management of ischemic and hemorrhagic stroke. We queried a United States Food and Drug Administration (FDA) database, along with PubMed and private company websites, to identify the recent literature assessing the clinical performance of FDA-approved AI/ML-enabled technologies. The FDA has approved 22 AI/ML-enabled technologies that triage brain imaging for more immediate diagnosis or promote post-stroke neurological/functional recovery. Technologies that assist with diagnosis predominantly use convolutional neural networks to identify abnormal brain images (e.g., CT perfusion). These technologies perform comparably to neuroradiologists, improve clinical workflows (e.g., time from scan acquisition to reading), and improve patient outcomes (e.g., days spent in the neurological ICU). Two devices are indicated for post-stroke rehabilitation by leveraging neuromodulation techniques. Multiple FDA-approved technologies exist that can help clinicians better diagnose and manage stroke. This review summarizes the most up-to-date literature regarding the functionality, performance, and utility of these technologies so clinicians can make informed decisions when using them in practice.

Identification of infarct core and ischemic penumbra using computed tomography perfusion and deep learning

Purpose

The size and location of infarct and penumbra are key to decision-making for acute ischemic stroke (AIS) management. CT perfusion (CTP) software estimate infarct and penumbra volume using contralateral hemisphere relative thresholding. This approach is not robust and widely contested by the scientific community. In this study, we investigate the use of deep learning-based algorithms to efficiently locate infarct and penumbra tissue on CTP hemodynamic maps.

Approach

CTP scans were retrospectively collected for 60 and 59 patients in the infarct only and infarct + penumbra substudies respectively. Commercial CTP software was used to generate cerebral blood flow, cerebral blood volume, mean transit time, time to peak, and delay time maps. U-Net-shaped architectures were trained to segment infarct or infarct + penumbra. Test-time-augmentation, ensembling, and watershed segmentation were used as postprocessing techniques. Segmentation performance was evaluated using Dice coefficients (DC) and mean absolute volume errors (MAVE).

Results

The algorithm segmented infarct tissue resulted in DC of 0.64 ± 0.03 (0.63, 0.65), and MAVE of 4.91 ± 0.94 (4.5, 5.32) mL. In comparison, the commercial software predicted infarct with a DC of 0.31 ± 0.17 (0.26, 0.36) and MAVE of 9.77 ± 8.35 (7.12, 12.42) mL. The algorithm was able to segment infarct + penumbra with a DC of 0.61 ± 0.04 (0.6, 0.63), and MAVE of 6.51 ± 1.37 (5.91, 7.11) mL. In comparison, the commercial software predicted infarct + penumbra with a DC of 0.3 ± 0.19 (0.25, 0.35) and MAVE of 9.18 ± 7.55 (7.25, 11.11) mL.

Conclusions

Use of deep learning algorithms to assess severity of AIS in terms of infarct and penumbra volume is precise and outperforms current relative thresholding methods. Such an algorithm would enhance the reliability of CTP in guiding treatment decisions.

Automatic identification of early ischemic lesions on non-contrast CT with deep learning approach

Early ischemic lesion on non-contrast computed tomogram (NCCT) in acute stroke can be subtle and need confirmation with magnetic resonance (MR) image for treatment decision-making. We retrospectively included the NCCT slices of 129 normal subjects and 546 ischemic stroke patients (onset < 12 h) with corresponding MR slices as reference standard from a prospective registry of Chang Gung Research Databank. In model selection, NCCT slices were preprocessed and fed into five different pre-trained convolutional neural network (CNN) models including Visual Geometry Group 16 (VGG16), Residual Networks 50, Inception-ResNet-v2, Inception-v3, and Inception-v4. In model derivation, the customized-VGG16 model could achieve an accuracy of 0.83, sensitivity 0.85, F-score 0.80, specificity 0.82, and AP 0.82 after using a tenfold cross-validation method, outperforming the pre-trained VGG16 model. In model evaluation, the customized-VGG16 model could correctly identify 53 in 58 subjects (91.37%) including 29 ischemic stroke patients and 24 normal subjects and reached the sensitivity of 86.95% in identifying ischemic NCCT slices (200/230), irrespective of supratentorial or infratentorial lesions. The customized-VGG16 CNN model can successfully identify the presence of early ischemic lesions on NCCT slices using the concept of automatic feature learning. Further study will be proceeded to detect the location of ischemic lesion.