Predicting intraventricular hemorrhage growth with a machine learning-based, radiomics-clinical model

A Comparative Study of a Nomogram and Machine Learning Models in Predicting Early Hematoma Expansion in Hypertensive Intracerebral Hemorrhage

RATIONALE AND OBJECTIVES

Early identification for hematoma expansion can help improve patient outcomes. Presently, there are many methods to predict hematoma expansion. This study compared a variety of models to find a model suitable for clinical promotion.

MATERIALS AND METHODS

Non-contrast head CT images and clinical data were collected from 203 patients diagnosed with hypertensive intracerebral hemorrhage. Radiomics features were extracted from all CT images, and the dataset was randomly divided into training and validation sets (7:3 ratio) after applying the synthetic minority oversampling method. The radiomics score (Radscore) was calculated using least absolute shrinkage and selection operator (LASSO) regression, combined with selected clinical predictors, to develop a nomogram and four machine learning (ML) models: logistic regression, random forest, support vector machine, and extreme gradient boosting (XGBoost). Discrimination, calibration and clinical usefulness of the nomogram and ML models were assessed.

RESULTS

The nomogram and ML models were integrated with Radscore and clinical predictors. The nomogram demonstrated favorable discriminatory ability in the training set with an AUC of 0.80, which was confirmed in the validation set (AUC=0.76). Among the ML models, the XGBoost model achieved the highest AUC (training set=0.89 and validation set=0.85), surpassing that of the nomogram. The XGBoost model exhibited good clinical usefulness.

CONCLUSION

Both the nomogram and ML models constructed by non-contrast head CT image-based Radscore integrated with clinical predictors can predict early hematoma expansion of hypertensive intracerebral hemorrhage, and the XGBoost model had the highest prediction performance and best clinical usefulness.

Machine learning for predicting hematoma expansion in spontaneous intracerebral hemorrhage: a systematic review and meta-analysis

PURPOSE

Early identification of hematoma enlargement and persistent hematoma expansion (HE) in patients with cerebral hemorrhage is increasingly crucial for determining clinical treatments. However, due to the lack of clinically effective tools, radiomics has been gradually introduced into the early identification of hematoma enlargement. Though, radiomics has limited predictive accuracy due to variations in procedures. Therefore, we conducted a systematic review and meta-analysis to explore the value of radiomics in the early detection of HE in patients with cerebral hemorrhage.

METHODS

Eligible studies were systematically searched in PubMed, Embase, Cochrane and Web of Science from inception to April 8, 2024. English articles are considered eligible. The radiomics quality scoring (RQS) tool was used to evaluate included studies.

RESULTS

A total of 34 studies were identified with sample sizes ranging from 108 to 3016. Eleven types of models were involved, and the types of modeling contained mainly clinical, radiomic, and radiomic plus clinical features. The radiomics models seem to have better performance (0.77 and 0.73 C-index in the training cohort and validation cohort, respectively) than the clinical models (0.69 C-index in the training cohort and 0.70 C-index in the validation cohort) in discriminating HE. However, the C-index was the highest for the combined model in both the training (0.82) and validation (0.79) cohorts.

CONCLUSIONS

Machine learning based on radiomic plus clinical features has the best predictive performance for HE, followed by machine learning based on radiomic features, and can be used as a potential tool to assist clinicians in early judgment.

Predicting severe intraventricular hemorrhage or early death using machine learning algorithms in VLBWI of the Korean Neonatal Network Database

Severe intraventricular hemorrhage (IVH) in premature infants can lead to serious neurological complications. This retrospective cohort study used the Korean Neonatal Network (KNN) dataset to develop prediction models for severe IVH or early death in very-low-birth-weight infants (VLBWIs) using machine-learning algorithms. The study included VLBWIs registered in the KNN database. The outcome was the diagnosis of IVH Grades 3-4 or death within one week of birth. Predictors were categorized into three groups based on their observed stage during the perinatal period. The dataset was divided into derivation and validation sets at an 8:2 ratio. Models were built using Logistic Regression with Ridge Regulation (LR), Random Forest, and eXtreme Gradient Boosting (XGB). Stage 1 models, based on predictors observed before birth, exhibited similar performance. Stage 2 models, based on predictors observed up to one hour after birth, showed improved performance in all models compared to Stage 1 models. Stage 3 models, based on predictors observed up to one week after birth, showed the best performance, particularly in the XGB model. Its integration into treatment and management protocols can potentially reduce the incidence of permanent brain injury caused by IVH during the early stages of birth.

An Online Dynamic Radiomics-Clinical Nomogram to Predict Recurrence in Patients with Spontaneous Intracerebral Hemorrhage

OBJECTIVE

Radiomics can reflect the heterogeneity within the focus. We aim to explore whether radiomics can predict recurrent intracerebral hemorrhage (RICH) and develop an online dynamic nomogram to predict it.

METHODS

This retrospective study collected the clinical and radiomics features of patients with spontaneous intracerebral hemorrhage seen in our hospital from October 2013 to October 2016. We used the minimum redundancy maximum relevancy and the least absolute shrinkage and selection operator methods to screen radiomics features and calculate the Rad-score. We use the univariate and multivariate analyses to screen clinical predictors. Optimal clinical features and Rad-score were used to construct different logistics regression models called the clinical model, radiomics model, and combined-logistic regression model. DeLong testing was performed to compare performance among different models. The model with the best predictive performance was used to construct an online dynamic nomogram.

RESULTS

Overall, 304 patients with intracerebral hemorrhage were enrolled in this study. Fourteen radiomics features were selected to calculate the Rad-score. The patients with RICH had a significantly higher Rad-score than those without (0.5 vs. -0.8; P< 0.001). The predictive performance of the combined-logistic regression model with Rad-score was better than that of the clinical model for both the training (area under the receiver operating curve, 0.81 vs. 0.71; P = 0.02) and testing (area under the receiver operating curve, 0.65 vs. 0.58; P = 0.04) cohorts statistically.

CONCLUSIONS

Radiomics features were determined related to RICH. Adding Rad-score into conventional clinical models significantly improves the prediction efficiency. We developed an online dynamic nomogram to accurately and conveniently evaluate RICH.

Prediction of Early Perihematomal Edema Expansion Based on Noncontrast Computed Tomography Radiomics and Machine Learning in Intracerebral Hemorrhage

OBJECTIVES

To investigate the predictive value of noncontrast computed tomography (NCCT) models based on radiomics features and machine learning for early perihematomal edema (PHE) expansion in patients with spontaneous intracerebral hemorrhage (ICH).

METHODS

We retrospectively reviewed NCCT data from 214 patients with spontaneous ICH. All radiomics features were extracted from volume of interest of hematomas on admission scans. A total of 8 machine learning methods were applied for constructing models in the training and the test set. Receiver operating characteristic analysis and the areas under the curve were used to evaluate the predictive value.

RESULTS

A total of 23 features were finally selected to establish models of early PHE expansion after feature screening. Patients were randomly assigned into training (n = 171) and test (n = 43) sets. The accuracy, sensitivity, and specificity in the test set were 72.1%, 90.0%, and 66.7% for the support vector machine model; 79.1%, 70.0%, and 84.4% for the k-nearest neighbor model; 88.4%, 90.0%, and 87.9% for the logistic regression model; 74.4%, 90.0%, and 69.7% for the extra tree model; 74.4%, 90.0%, and 69.7% for the extreme gradient boosting model; 83.7%, 100%, and 78.8% for the multilayer perceptron (MLP) model; 72.1%, 100%, and 65.6% for the light gradient boosting machine model; and 60.5%, 90.0%, and 53.1% for the random forest model, respectively.

CONCLUSIONS

The MLP model seemed to be the best model for prediction of PHE expansion in patients with ICH. NCCT models based on radiomics features and machine learning could predict early PHE expansion and improve the discrimination of identify spontaneous intracerebral hemorrhage patients at risk of early PHE expansion.

Difference of mean Hounsfield units (dHU) between follow-up and initial noncontrast CT scan predicts 90-day poor outcome in spontaneous supratentorial acute intracerebral hemorrhage with deep convolutional neural networks

OBJECTIVES

This study aimed to investigate the usefulness of a new non-contrast CT scan (NCCT) sign called the dHU, which represented the difference in mean Hounsfield unit values between follow-up and the initial NCCT for predicting 90-day poor functional outcomes in acute supratentorial spontaneous intracerebral hemorrhage(sICH) using deep convolutional neural networks.

METHODS

A total of 377 consecutive patients with sICH from center 1 and 91 patients from center 2 (external validation set) were included. A receiver operating characteristic (ROC) analysis was performed to determine the critical value of dHU for predicting poor outcome at 90 days. Modified Rankin score (mRS) >3 or >2 was defined as the primary and secondary poor outcome, respectively. Two multivariate models were developed to test whether dHU was an independent predictor of the two unfavorable functional outcomes.

RESULTS

The ROC analysis showed that a dHU >2.5 was a critical value to predict the poor outcomes (mRS >3) in sICH. The sensitivity, specificity, and accuracy of dHU >2.5 for poor outcome prediction were 37.5%, 86.0%, and 70.6%, respectively. In multivariate models developed after adjusting for all elements of the ICH score and hematoma expansion, dHU >2.5 was an independent predictor of both primary and secondary poor outcomes (OR = 2.61, 95% CI [1.32,5.13], P = 0.006; OR = 2.63, 95% CI [1.36,5.10], P = 0.004, respectively). After adjustment for all possible significant predictors (p < 0.05) by univariate analysis, dHU >2.5 had a positive association with primary and secondary poor outcomes (OR = 3.25, 95% CI [1.52,6.98], P = 0.002; OR = 3.42, 95% CI [1.64,7.15], P = 0.001).

CONCLUSIONS

The dHU of hematoma based on serial CT scans is independently associated with poor outcomes after acute sICH, which may help predict clinical evolution and guide therapy for sICH patients.

The Added Value of Intraventricular Hemorrhage on the Radiomics Analysis for the Prediction of Hematoma Expansion of Spontaneous Intracerebral Hemorrhage

Background: Among patients undergoing head computed tomography (CT) scans within 3 h of spontaneous intracerebral hemorrhage (sICH), 28% to 38% have hematoma expansion (HE) on follow-up CT. This study aimed to predict HE using radiomics analysis and investigate the impact of intraventricular hemorrhage (IVH) compared with the conventional approach based on intraparenchymal hemorrhage (IPH) alone. Methods: This retrospective study enrolled 127 patients with baseline and follow-up non-contrast CT (NCCT) within 4~72 h of sICH. IPH and IVH were outlined separately for performing radiomics analysis. HE was defined as an absolute hematoma growth > 6 mL or percentage growth > 33% of either IPH (HEP) or a combination of IPH and IVH (HEP+V) at follow-up. Radiomic features were extracted using PyRadiomics, and then the support vector machine (SVM) was used to build the classification model. For each case, a radiomics score was generated to indicate the probability of HE. Results: There were 57 (44.9%) HEP and 70 (55.1%) non-HEP based on IPH alone, and 58 (45.7%) HEP+V and 69 (54.3%) non-HEP+V based on IPH + IVH. The majority (>94%) of HE patients had poor early outcomes (death or modified Rankin Scale > 3 at discharge). The radiomics model built using baseline IPH to predict HEP (RMP) showed 76.4% accuracy and 0.73 area under the ROC curve (AUC). The other model using IPH + IVH to predict HEP+V (RMP+V) had higher accuracy (81.9%) with AUC = 0.80, and this model could predict poor outcomes. The sensitivity/specificity of RMP and RMP+V for HE prediction were 71.9%/80.0% and 79.3%/84.1%, respectively. Conclusion: The proposed radiomics approach with additional IVH information can improve the accuracy in prediction of HE, which is associated with poor clinical outcomes. A reliable radiomics model may provide a robust tool to help manage ICH patients and to enroll high-risk ICH cases into anti-expansion or neuroprotection drug trials.

Nomogram based on ultrasound radiomics score and clinical variables for predicting histologic subtypes of epithelial ovarian cancer

OBJECTIVE

Ovarian cancer is one of the most common causes of death in gynecological tumors, and its most common type is epithelial ovarian cancer (EOC). This study aimed to establish a radiomics signature based on ultrasound images to predict the histopathological types of EOC.

METHODS

Overall, 265 patients with EOC who underwent preoperative ultrasonography and surgery were eligible. They were randomly sorted into two cohorts (training cohort: test cohort = 7:3). We outlined the region of interest of the tumor on the ultrasound images of the lesion. Then, the radiomics features were extracted. Clinical, Rad-score and combined models were constructed based on the least absolute shrinkage, selection operator, and logistic regression analysis. The performance of the models was evaluated using receiver operating characteristic curves and decision curve analysis (DCA). A nomogram was formulated based on the combined prediction model.

RESULTS

The combined model had good performance in predicting EOC histopathological types, with an AUC of 0.83 (95% CI: 0.77-0.90) and 0.82 (95% CI: 0.71-0.93) in the training and test cohorts, respectively. The calibration curves showed that the nomogram estimation was consistent with the actual observations. DCA also verified the clinical value of the combined model.

CONCLUSIONS

The combined model containing clinical and ultrasound radiomics features showed an excellent performance in predicting type I and type II EOC.

ADVANCES IN KNOWLEDGE

This study presents the first application of ultrasound radiomics features to distinguish EOC histopathological types. The proposed clinical-radiomics nomogram could help gynecologists non-invasively identify EOC types before surgery.

Moderate-Severe White Matter Lesion Predicts Delayed Intraventricular Hemorrhage in Intracerebral Hemorrhage

BACKGROUND

Most existing studies have focused on the correlation between white matter lesion (WML) and baseline intraventricular hemorrhage (IVH) in patients with intracerebral hemorrhage (ICH), whereas few studies have investigated the relationship between WML severity and delayed IVH after admission. This study aimed to investigate the correlation between WML severity and delayed IVH and to verify the association between WML and baseline IVH.

METHODS

A total of 480 patients with spontaneous ICH from February 2018 to October 2020 were selected. WML was scored using the Van Swieten Scale, with scores of 0-2 representing nonslight WML and scores of 3-4 representing moderate-severe WML. We determined the presence of IVH on baseline (< 6 h) and follow-up computed tomography (< 72 h) images. Univariate analysis and multiple logistic regression were used to analyze the influencing factors of baseline and delayed IVH.

RESULTS

Among 480 patients with ICH, 172 (35.8%) had baseline IVH, and there was a higher proportion of moderate-severe WML in patients with baseline IVH (20.3%) than in those without baseline IVH (12.7%) (P = 0.025). Among 308 patients without baseline IVH, delayed IVH was found in 40 patients (12.9%), whose proportion of moderate-severe WML (25.0%) was higher than that in patients without delayed IVH (10.8%) (P = 0.012). Multiple logistic regression results showed that moderate-severe WML was independently correlated with baseline IVH (P = 0.006, odds ratio = 2.266, 95% confidence interval = 1.270-4.042) and delayed IVH (P = 0.002, odds ratio = 7.009, 95% confidence interval = 12.086-23.552).

CONCLUSIONS

Moderate-severe WML was an independent risk factor for delayed IVH as well as baseline IVH.

Machine learning model prediction of 6-month functional outcome in elderly patients with intracerebral hemorrhage

Spontaneous intracerebral hemorrhage (ICH) has an increasing incidence and a worse outcome in elderly patients. The ability to predict the functional outcome in these patients can be helpful in supporting treatment decisions and establishing prognostic expectations. We evaluated the performance of a machine learning (ML) model to predict the 6-month functional status in elderly patients with ICH leveraging the predictive value of the clinical characteristics at hospital admission. Data were extracted by a retrospective multicentric database of patients ≥ 70 years of age consecutively admitted for the management of spontaneous ICH between January 1, 2014 and December 31, 2019. Relevant demographic, clinical, and radiological variables were selected by a feature selection algorithm (Boruta) and used to build a ML model. Outcome was determined according to the Glasgow Outcome Scale (GOS) at 6 months from ICH: dead (GOS 1), poor outcome (GOS 2–3: vegetative status/severe disability), and good outcome (GOS 4–5: moderate disability/good recovery). Ten features were selected by Boruta with the following relative importance order in the ML model: Glasgow Coma Scale, Charlson Comorbidity Index, ICH score, ICH volume, pupillary status, brainstem location, age, anticoagulant/antiplatelet agents, intraventricular hemorrhage, and cerebellar location. Random forest prediction model, evaluated on the hold-out test set, achieved an AUC of 0.96 (0.94–0.98), 0.89 (0.86–0.93), and 0.93 (0.90–0.95) for dead, poor, and good outcome classes, respectively, demonstrating high discriminative ability. A random forest classifier was successfully trained and internally validated to stratify elderly patients with spontaneous ICH into prognostic subclasses. The predictive value is enhanced by the ability of ML model to identify synergy among variables.

A data-driven health index for neonatal morbidities

Whereas prematurity is a major cause of neonatal mortality, morbidity, and lifelong impairment, the degree of prematurity is usually defined by the gestational age (GA) at delivery rather than by neonatal morbidity. Here we propose a multi-task deep neural network model that simultaneously predicts twelve neonatal morbidities, as the basis for a new data-driven approach to define prematurity. Maternal demographics, medical history, obstetrical complications, and prenatal fetal findings were obtained from linked birth certificates and maternal/infant hospitalization records for 11,594,786 livebirths in California from 1991 to 2012. Overall, our model outperformed traditional models to assess prematurity which are based on GA and/or birthweight (area under the precision-recall curve was 0.326 for our model, 0.229 for GA, and 0.156 for small for GA). These findings highlight the potential of using machine learning techniques to predict multiple prematurity phenotypes and inform clinical decisions to prevent, diagnose and treat neonatal morbidities.

•

Traditional definitions of prematurity based on gestational age need to be updated

•

Deep learning of maternal clinical data improves predictions of neonatal morbidity

•

Data-driven model leverages birthweight, type of delivery and maternal race

•

Accurate risk prediction can inform clinical decisions

CT-based radiomics for differentiating intracranial contrast extravasation from intraparenchymal haemorrhage after mechanical thrombectomy

OBJECTIVE

To develop a nonenhanced CT-based radiomic signature for the differentiation of iodinated contrast extravasation from intraparenchymal haemorrhage (IPH) following mechanical thrombectomy.

METHODS

Patients diagnosed with acute ischaemic stroke who underwent mechanical thrombectomy in 4 institutions from December 2017 to June 2020 were included in this retrospective study. The study population was divided into a training cohort and a validation cohort. The nonenhanced CT images taken after mechanical thrombectomy were used to extract radiomic features. The maximum relevance minimum redundancy (mRMR) algorithm was used to eliminate confounding variables. Afterwards, least absolute shrinkage and selection operator (LASSO) logistic regression was used to generate the radiomic signature. The diagnostic performance of the radiomic signature was evaluated by the area under the curve (AUC), accuracy, specificity, sensitivity, positive predictive value (PPV), and negative predictive value (NPV).

RESULTS

A total of 166 intraparenchymal areas of hyperattenuation from 101 patients were used. The areas of hyperattenuation were randomly allocated to the training and validation cohorts at a ratio of 7:3. The AUC of the radiomic signature was 0.848 (95% confidence interval (CI) 0.780-0.917) in the training cohort and 0.826 (95% CI 0.705-0.948) in the validation cohort. The accuracy of the radiomic signature was 77.6%, with a sensitivity of 76.7%, a specificity of 78.9%, a PPV of 85.2%, and a NPV of 68.2% in the validation cohort.

CONCLUSIONS

The radiomic signature constructed based on initial post-operative nonenhanced CT after mechanical thrombectomy can effectively differentiate IPH from iodinated contrast extravasation.

KEY POINTS

• Radiomic features were extracted from intraparenchymal areas of hyperattenuation on initial post-operative CT scans after mechanical thrombectomy. • The nonenhanced CT-based radiomic signature can differentiate IPH from iodinated contrast extravasation early. • The radiomic signature may help prevent unnecessary rescanning after mechanical thrombectomy, especially in cases where contrast extravasation is highly suggestive.

Ultrasound-based radiomics score: a potential biomarker for the prediction of progression-free survival in ovarian epithelial cancer

PURPOSE

More than 80% of patients with ovarian epithelial cancer (OEC) show complete remission after initial treatment but eventually experience recurrence of the disease. This study aimed to develop a radiomics signature to identify a new prognostic indicator based on preoperative ultrasound imaging.

METHODS

A total of 111 patients with OEC who underwent transvaginal ultrasound before surgery were included. Of these, 76 were divided into the training cohort and 35 into the test cohort. We defined the region of interest (ROI) of the tumor by manually drawing the tumor contour on the ultrasound image of the lesion. The radiomics features were extracted from ultrasound images. The radiomics score (Rad-Score) was constructed using the least absolute shrinkage and selection operator (LASSO) analysis and Cox regression. Combined with the ultrasound radiomics features, significant clinical variables were also used to establish predictive models for 5-year progression-free survival (PFS) prediction. The efficiency of the model was evaluated using the area under the curve (AUC). Kaplan-Meier analysis was used to evaluate the association between the Rad-Score and PFS.

RESULTS

The combined model was superior to the clinical and Rad-Score models in estimating 5-year PFS and achieved an AUC of 0.868 (95%CI 0.766-0.971) in the training cohort. The Rad-Score was negatively correlated with prognosis in the training and test cohorts.

CONCLUSIONS

The combined model that incorporated both clinical parameters and ultrasound radiomics features achieved a good prognosis in patients with OEC, which might aid clinical decision-making.

Classifying Ruptured Middle Cerebral Artery Aneurysms With a Machine Learning Based, Radiomics-Morphological Model: A Multicentral Study

Objective

Radiomics and morphological features were associated with aneurysms rupture. However, the multicentral study of their predictive power for specific-located aneurysms rupture is rare. We aimed to determine robust radiomics features related to middle cerebral artery (MCA) aneurysms rupture and evaluate the additional value of combining morphological and radiomics features in the classification of ruptured MCA aneurysms.

Methods

A total of 632 patients with 668 MCA aneurysms (423 ruptured aneurysms) from five hospitals were included. Radiomics and morphological features of aneurysms were extracted on computed tomography angiography images. The model was developed using a training dataset (407 patients) and validated with the internal (152 patients) and external validation (73 patients) datasets. The support vector machine method was applied for model construction. Optimal radiomics, morphological, and clinical features were used to develop the radiomics model (R-model), morphological model (M-model), radiomics-morphological model (RM-model), clinical-morphological model (CM-model), and clinical-radiomics-morphological model (CRM-model), respectively. A comprehensive nomogram integrating clinical, morphological, and radiomics predictors was generated.

Results

We found seven radiomics features and four morphological predictors of MCA aneurysms rupture. The R-model obtained an area under the receiver operating curve (AUC) of 0.822 (95% CI, 0.776, 0.867), 0.817 (95% CI, 0.744, 0.890), and 0.691 (95% CI, 0.567, 0.816) in the training, temporal validation, and external validation datasets, respectively. The RM-model showed an AUC of 0.848 (95% CI, 0.810, 0.885), 0.865 (95% CI, 0.807, 0.924), and 0.721 (95% CI, 0.601, 0.841) in the three datasets. The CRM-model obtained an AUC of 0.856 (95% CI, 0.820, 0.892), 0.882 (95% CI, 0.828, 0.936), and 0.738 (95% CI, 0.618, 0.857) in the three datasets. The CRM-model and RM-model outperformed the CM-model and M-model in the internal datasets (p < 0.05), respectively. But these differences were not statistically significant in the external dataset. Decision curve analysis indicated that the CRM-model obtained the highest net benefit for most of the threshold probabilities.

Conclusion

Robust radiomics features were determined related to MCA aneurysm rupture. The RM-model exhibited good ability in classifying ruptured MCA aneurysms. Integrating radiomics features into conventional models might provide additional value in ruptured MCA aneurysms classification.