Predicting the clinical prognosis of acute ischemic stroke using machine learning: an application of radiomic biomarkers on non-contrast CT after intravascular interventional treatment

 Predicting failure to recover is ABSOLUTELY FUCKING USELESS! When the hell will you do the research that produces recovery? Maybe when you are the 1 in 4 per WHO that has a stroke?

Predicting the clinical prognosis of acute ischemic stroke using machine learning: an application of radiomic biomarkers on non-contrast CT after intravascular interventional treatment

Hongxian Gu¹Yuting Yan²Xiaodong He²Yuyun Xu²Yuguo Wei³Yuan Shao^2*

• ¹Department of Radiology, The People's Hospital of Jianyang City, Jianyang, Sichuan Province, China
• ²Center for Rehabilitation Medicine, Department of Radiology, Zhejiang Provincial People's Hospital, Hangzhou, Zhejiang, China
• ³GE Healthcare Life Sciences, Hangzhou, Jiangsu, China

Purpose: This study aimed to develop a radiomic model based on non-contrast computed tomography (NCCT) after interventional treatment to predict the clinical prognosis of acute ischemic stroke (AIS) with large vessel occlusion.

Methods: We retrospectively collected 141 cases of AIS from 2016 to 2020 and analyzed the patients' clinical data as well as NCCT data after interventional treatment. Then, the total dataset was divided into training and testing sets according to the subject serial number. The cerebral hemispheres on the infarct side were segmented for radiomics signature extraction. After radiomics signatures were standardized and dimensionality reduced, the training set was used to construct a radiomics model using machine learning. The testing set was then used to validate the prediction model, which was evaluated based on discrimination, calibration, and clinical utility. Finally, a joint model was constructed by incorporating the radiomics signatures and clinical data.

Results: The AUCs of the joint model, radiomics signature, NIHSS score, and hypertension were 0.900, 0.863, 0.727, and 0.591, respectively, in the training set. In the testing set, the AUCs of the joint model, radiomics signature, NIHSS score, and hypertension were 0.885, 0.840, 0.721, and 0.590, respectively.

Conclusion: Our results provided evidence that using post-interventional NCCT for a radiomic model could be a valuable tool in predicting the clinical prognosis of AIS with large vessel occlusion.(So you're going to tell your patients they are not going to recover because you incompetently never did the research on getting survivors recovered? Good luck listening to the screaming.)

Introduction

AIS is a neurological emergency with high rates of disability and mortality (Regenhardt et al., 2018). According to statistics, ~25–35% of strokes manifest as large vessel occlusion, and this group is the main target for intravascular interventional therapy (Kidwell et al., 2013). However, the hyperdense areas on postoperative NCCT often confuse clinicians as to whether it was a hemorrhage or contrast agent and affect subsequent treatment and clinical prognosis.

The relationship between the hyperdense area and clinical outcomes remains uncertain. Some studies have shown that patients with the hyperdense area had a higher score on the modified Rankin Scale (mRS) score at discharge or 90 days than those without the hyperdense area (Payabvash et al., 2014, 2015; Rouchaud et al., 2014; Chen et al., 2019, 2020), while others indicated that it did not affect functional outcomes (Lummel et al., 2014; An et al., 2019). We would like to use a new machine learning tool that could obtain more information, including the area of the hyperdense area, the area of concomitant hypodense infarction, the histogram of CT value distribution, and the degree of brain parenchyma swelling to make a one-stop prediction of clinical outcomes.

Radiomics, as a new technology, transforms subjective visual interpretation into image data-driven objective evaluation in a non-invasive way. It can extract a large number of quantitative features, such as shape, intensity, and texture, from images and further reflect more biological information related to the disease (Lambin et al., 2012; Yip and Aerts, 2016; Avanzo et al., 2017). Radiomics has successfully demonstrated the potential for multiple applications in stroke, and the extracted features can be used to diagnose stroke lesions, predict early transformation, and assess the long-term prognosis after stroke onset (Chen et al., 2021; Jiang et al., 2021). Peter et al. (2017) identified six texture features from NCCT images that could differentiate ischemic lesions from their contralateral normal tissues. In addition, Tang et al. (2020) quantified the penumbra and core area from both the apparent diffusion coefficient and cerebral blood flow maps in patients with AIS (< 9 h) using radiomic analysis, and in the external dataset, the constructed radiomic nomogram could strongly predict favorable clinical outcomes at 7 days and 3 months. Clinically, NCCT is the first choice for AIS patients after intervention because it is efficient, non-invasive, and low in cost. Nevertheless, little is known about the relationship between the radiomics signatures based on NCCT after AIS intervention and the clinical prognosis.

Therefore, we aimed to develop a radiomics model to predict the clinical prognosis of AIS patients with interventional treatment. Then, the correlation between texture features and clinical outcome was further elucidated to identify potential biomarkers for clinical prognosis.

More at link.