Current clinical practice relies on clinical history to determine the time since stroke (TSS) onset. Imaging-based determination of acute stroke onset time could provide critical information to clinicians in deciding stroke treatment options, such as thrombolysis. The patients with unknown or unwitnessed TSS are usually excluded from thrombolysis, even if their symptoms began within the therapeutic window. In this paper, we demonstrate a machine learning approach for TSS classification using routinely acquired imaging sequences. We develop imaging features from the magnetic resonance (MR) images and train machine learning models to classify the TSS. We also propose a deep-learning model to extract hidden representations for the MR perfusion-weighted images and demonstrate classification improvement by incorporating these additional deep features. The cross-validation results show that our best classifier achieved an area under the curve of 0.765, with a sensitivity of 0.788 and a negative predictive value of 0.609, outperforming existing methods. We show that the features generated by our deep-learning algorithm correlate with the MR imaging features, and validate the robustness of the model on imaging parameter variations (e.g., year of imaging). This paper advances magnetic resonance imaging analysis one-step-closer to an operational decision support tool for stroke treatment guidance.

CT is crucial for diagnosing chest diseases, with image quality affected by spatial resolution. Thick-slice CT remains prevalent in practice due to cost considerations, yet its coarse spatial resolution may hinder accurate diagnoses. Our multicenter study develops a deep learning synthetic model with Convolutional-Transformer hybrid encoder-decoder architecture for generating thin-slice CT from thick-slice CT on a single center (1576 participants) and access the synthetic CT on three cross-regional centers (1228 participants). The qualitative image quality of synthetic and real thin-slice CT is comparable (p = 0.16). Four radiologists’ accuracy in diagnosing community-acquired pneumonia using synthetic thin-slice CT surpasses thick-slice CT (p < 0.05), and matches real thin-slice CT (p > 0.99). For lung nodule detection, sensitivity with thin-slice CT outperforms thick-slice CT (p < 0.001) and comparable to real thin-slice CT (p > 0.05). These findings indicate the potential of our model to generate high-quality synthetic thin-slice CT as a practical alternative when real thin-slice CT is preferred but unavailable.

Background: Cerebral venous thrombosis (CVT) is a rare cerebrovascular disease. Routine brain magnetic resonance imaging is commonly used to diagnose CVT. This study aimed to develop and evaluate a novel deep learning (DL) algorithm for detecting CVT using routine brain magnetic resonance imaging. Methods: Routine brain magnetic resonance imaging, including T1-weighted, T2-weighted, and fluid-attenuated inversion recovery images of patients suspected of CVT from April 2014 through December 2019 who were enrolled from a CVT registry, were collected. The images were divided into 2 data sets: a development set and a test set. Different DL algorithms were constructed in the development set using 5-fold cross-validation. Four radiologists with various levels of expertise independently read the images and performed diagnosis within the test set. The diagnostic performance on per-patient and per-segment diagnosis levels of the DL algorithms and radiologist’s assessment were evaluated and compared. Results: A total of 392 patients, including 294 patients with CVT (37±14 years, 151 women) and 98 patients without CVT (42±15 years, 65 women), were enrolled. Of these, 100 patients (50 CVT and 50 non-CVT) were randomly assigned to the test set, and the other 292 patients comprised the development set. In the test set, the optimal DL algorithm (multisequence multitask deep learning algorithm) achieved an area under the curve of 0.96, with a sensitivity of 96% (48/50) and a specificity of 88% (44/50) on per-patient diagnosis level, as well as a sensitivity of 88% (129/146) and a specificity of 80% (521/654) on per-segment diagnosis level. Compared with 4 radiologists, multisequence multitask deep learning algorithm showed higher sensitivity both on per-patient (all P <0.05) and per-segment diagnosis levels (all P <0.001). Conclusions: The CVT-detected DL algorithm herein improved diagnostic performance of routine brain magnetic resonance imaging, with high sensitivity and specificity, which provides a promising approach for detecting CVT.

Purpose: The T2-FLAIR mismatch sign has shown promise in determining IDH mutant 1p/19q non-co-deleted gliomas with a high specificity and modest sensitivity. To develop a multi-parametric radiomic model using MRI to predict 1p/19q co-deletion status in patients with newly diagnosed IDH1 mutant glioma and to perform a comparative analysis to T2-FLAIR mismatch sign+. Methods: In this retrospective study, patients with diagnosis of IDH1 mutant gliomas with known 1p/19q status who had preoperative MRI were included. T2-FLAIR mismatch was evaluated independently by two board-certified neuroradiologists. Texture features were extracted from glioma segmentation of FLAIR images. eXtremeGradient Boosting (XGboost) classifiers were used for model development. Leave-one-out-cross-validation (LOOCV) and external validation performances were reported for both the training and external validation sets. Results: A total of 103 patients were included for model development and 18 patients for external testing validation. The diagnostic performance (sensitivity/specificity/accuracy) in the determination of the 1p/19q co-deletion status was 59%/83%/67% (training) and 62.5%/70.0%/66.3% (testing) for the T2-FLAIR mismatch sign. This was significantly improved (p = 0.04) using the radiomics model to 77.9%/82.8%/80.3% (training) and 87.5%/89.9%/88.8% (testing), respectively. The addition of radiomics as a computer-assisted tool resulted in significant (p = 0.02) improvement in the performance of the neuroradiologist with 13 additional corrected cases in comparison to just using the T2-FLAIR mismatch sign. Conclusion: The proposed radiomic model provides much needed sensitivity to the highly specific T2-FLAIR mismatch sign in the determination of the 1p/19q non-co-deletion status and improves the overall diagnostic performance of neuroradiologists when used as an assistive tool.