The prediction of malignant middle cerebral artery infarction: a predicting approach using random forest

Understanding overfitting in random forest for probability estimation: a visualization and simulation study

BACKGROUND

Random forests have become popular for clinical risk prediction modeling. In a case study on predicting ovarian malignancy, we observed training AUCs close to 1. Although this suggests overfitting, performance was competitive on test data. We aimed to understand the behavior of random forests for probability estimation by (1) visualizing data space in three real-world case studies and (2) a simulation study.

METHODS

For the case studies, multinomial risk estimates were visualized using heatmaps in a 2-dimensional subspace. The simulation study included 48 logistic data-generating mechanisms (DGM), varying the predictor distribution, the number of predictors, the correlation between predictors, the true AUC, and the strength of true predictors. For each DGM, 1000 training datasets of size 200 or 4000 with binary outcomes were simulated, and random forest models were trained with minimum node size 2 or 20 using the ranger R package, resulting in 192 scenarios in total. Model performance was evaluated on large test datasets (N = 100,000).

RESULTS

The visualizations suggested that the model learned "spikes of probability" around events in the training set. A cluster of events created a bigger peak or plateau (signal), isolated events local peaks (noise). In the simulation study, median training AUCs were between 0.97 and 1 unless there were 4 binary predictors or 16 binary predictors with a minimum node size of 20. The median discrimination loss, i.e., the difference between the median test AUC and the true AUC, was 0.025 (range 0.00 to 0.13). Median training AUCs had Spearman correlations of around 0.70 with discrimination loss. Median test AUCs were higher with higher events per variable, higher minimum node size, and binary predictors. Median training calibration slopes were always above 1 and were not correlated with median test slopes across scenarios (Spearman correlation - 0.11). Median test slopes were higher with higher true AUC, higher minimum node size, and higher sample size.

CONCLUSIONS

Random forests learn local probability peaks that often yield near perfect training AUCs without strongly affecting AUCs on test data. When the aim is probability estimation, the simulation results go against the common recommendation to use fully grown trees in random forest models.

Comparison review of image classification techniques for early diagnosis of diabetic retinopathy

Diabetic retinopathy (DR) is one of the leading causes of vision loss in adults and is one of the detrimental side effects of the mass prevalence of Diabetes Mellitus (DM). It is crucial to have an efficient screening method for early diagnosis of DR to prevent vision loss. This paper compares and analyzes the various Machine Learning (ML) techniques, from traditional ML to advanced Deep Learning models. We compared and analyzed the efficacy of Convolutional Neural Networks (CNNs), Capsule Networks (CapsNet), K-Nearest Neighbor (KNN), Support Vector Machine (SVM), decision trees, and Random Forests. This paper also considers determining factors in the evaluation, including contrast enhancements, noise reduction, grayscaling, etc We analyze recent research studies and compare methodologies and metrics, including accuracy, precision, sensitivity, and specificity. The findings highlight the advanced performance of Deep Learning (DL) models, with CapsNet achieving a remarkable accuracy of up to 97.98% and a high precision rate, outperforming other traditional ML methods. The Contrast Limited Adaptive Histogram Equalization (CLAHE) preprocessing technique substantially enhanced the model's efficiency. Each ML method's computational requirements are also considered. While most advanced deep learning methods performed better according to the metrics, they are more computationally complex, requiring more resources and data input. We also discussed how datasets like MESSIDOR could be more straightforward and contribute to highly evaluated performance and that there is a lack of consistency regarding benchmark datasets across papers in the field. Using the DL models facilitates accurate early detection for DR screening, can potentially reduce vision loss risks, and improves accessibility and cost-efficiency of eye screening. Further research is recommended to extend our findings by building models with public datasets, experimenting with ensembles of DL and traditional ML models, and considering testing high-performing models like CapsNet.

A new machine learning model to predict the prognosis of cardiogenic brain infarction

Cardiogenic cerebral infarction (CCI) is a disease in which the blood supply to the blood vessels in the brain is insufficient due to atherosclerosis or stenosis of the coronary arteries in the patient's heart, which leads to neurological deficits. To predict the pathogenic factors of cardiogenic cerebral infarction, this paper proposes a machine learning based analytical prediction model. 494 patients with CCI who were hospitalized for the first time were consecutively included in the study between January 2017 and December 2021, and followed up every three months for one year after hospital discharge. Clinical, laboratory and imaging data were collected, and predictors associated with relapse and death in CCI patients at six months and one year after discharge were analyzed using univariate and multivariate logistic regression methods, meanwhile established a new machine learning model based on the enhanced moth-flame optimization (FTSAMFO) and the fuzzy K-nearest neighbor (FKNN), called BITSAMFO-FKNN, which is practiced on the dataset related to patients with CCI. Specifically, this paper proposes the spatial transformation strategy to increase the exploitation capability of moth-flame optimization (MFO) and combines it with the tree seed algorithm (TSA) to increase the search capability of MFO. In the benchmark function experiments FTSAMFO beat 5 classical algorithms and 5 recent variants. In the feature selection experiment, ten times ten-fold cross-validation trials showed that the BITSAMFO-FKNN model proved actual medical importance and efficacy, with an accuracy value of 96.61%, sensitivity value of 0.8947, MCC value of 0.9231, and F-Measure of 0.9444. The results of the trial showed that hemorrhagic conversion and lower LVDD/LVSD were independent risk factors for recurrence and death in patients with CCI. The established BITSAMFO-FKNN method is helpful for CCI prognosis and deserves further clinical validation.

Predicting who has delayed cerebral ischemia after aneurysmal subarachnoid hemorrhage using machine learning approach: a multicenter, retrospective cohort study

BACKGROUND

Early prediction of delayed cerebral ischemia (DCI) is critical to improving the prognosis of aneurysmal subarachnoid hemorrhage (aSAH). Machine learning (ML) algorithms can learn from intricate information unbiasedly and facilitate the early identification of clinical outcomes. This study aimed to construct and compare the ability of different ML models to predict DCI after aSAH. Then, we identified and analyzed the essential risk of DCI occurrence by preoperative clinical scores and postoperative laboratory test results.

METHODS

This was a multicenter, retrospective cohort study. A total of 1039 post-operation patients with aSAH were finally included from three hospitals in China. The training group contained 919 patients, and the test group comprised 120 patients. We used five popular machine-learning algorithms to construct the models. The area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, precision, and f1 score were used to evaluate and compare the five models. Finally, we performed a Shapley Additive exPlanations analysis for the model with the best performance and significance analysis for each feature.

RESULTS

A total of 239 patients with aSAH (23.003%) developed DCI after the operation. Our results showed that in the test cohort, Random Forest (RF) had an AUC of 0.79, which was better than other models. The five most important features for predicting DCI in the RF model were the admitted modified Rankin Scale, D-Dimer, intracranial parenchymal hematoma, neutrophil/lymphocyte ratio, and Fisher score. Interestingly, clamping or embolization for the aneurysm treatment was the fourth button-down risk factor in the ML model.

CONCLUSIONS

In this multicenter study, we compared five ML methods, among which RF performed the best in DCI prediction. In addition, the essential risks were identified to help clinicians monitor the patients at high risk for DCI more precisely and facilitate timely intervention.

An Explainable Artificial Intelligence Model to Predict Malignant Cerebral Edema after Acute Anterior Circulating Large-Hemisphere Infarction

INTRODUCTION

Malignant cerebral edema (MCE) is a serious complication and the main cause of poor prognosis in patients with large-hemisphere infarction (LHI). Therefore, the rapid and accurate identification of potential patients with MCE is essential for timely therapy. This study utilized an artificial intelligence-based machine learning approach to establish an interpretable model for predicting MCE in patients with LHI.

METHODS

This study included 314 patients with LHI not undergoing recanalization therapy. The patients were divided into MCE and non-MCE groups, and the eXtreme Gradient Boosting (XGBoost) model was developed. A confusion matrix was used to measure the prediction performance of the XGBoost model. We also utilized the SHapley Additive exPlanations (SHAP) method to explain the XGBoost model. Decision curve and receiver operating characteristic curve analyses were performed to evaluate the net benefits of the model.

RESULTS

MCE was observed in 121 (38.5%) of the 314 patients with LHI. The model showed excellent predictive performance, with an area under the curve of 0.916. The SHAP method revealed the top 10 predictive variables of the MCE such as ASPECTS score, NIHSS score, CS score, APACHE II score, HbA1c, AF, NLR, PLT, GCS, and age based on their importance ranking.

CONCLUSION

An interpretable predictive model can increase transparency and help doctors accurately predict the occurrence of MCE in LHI patients not undergoing recanalization therapy within 48 h of onset, providing patients with better treatment strategies and enabling optimal resource allocation.

Improvement in Midline Shift Is a Positive Prognostic Predictor for Malignant Middle Cerebral Artery Infarction Patients Undergoing Decompressive Craniectomy

Objective: The aim of this retrospective study is to evaluate the risk factors of malignant middle cerebral artery infarction (MMCAI) patients and explore an applicable prognostic predictor for MMCAI patients undergoing decompressive craniectomy (DC). Methods: Clinical data from the period 2012-2017 were retrospectively evaluated. Forty-three consecutive MMCAI patients undergoing DC were enrolled in this study. The 30-day mortality was assessed, and age, location, hypertension, pupil dilation, onset to operation duration, midline shift, and Glasgow Coma Scale (GCS) score were identified by univariate analysis and binary logistic regression. Results: In this retrospective study for DC patients, the 30-day mortality was 44.2%. In the univariate analysis, advanced age (≥60 years), right hemispheric location, hypertension, pupil dilation, shorter onset to operation duration (<48 h), improved midline shift (t = 4.214, p < 0.01), and lower pre-operation GCS score were significant predictors of death within 30 days. In binary logistic regression analysis, age [odds ratio (OR) = 1.141, 95% CI 1.011-1.287], the improvement of the midline shift (OR = 0.764, 95% CI 0.59-0.988), and pupillary dilation (OR = 15.10, 95% CI 1.374-165.954) were independent influencing factors. For the receiver operating characteristic (ROC) analysis of the relationship between post-operation outcomes and midline shift improvement, the area under the curve (AUC) was 0.844, and the cutoff point of midline shift improvement was 0.83 cm. Conclusion: Improved midline shift was a significant predictor of 30-day mortality. The improved midline shift of >0.83 cm indicated survival at 30 days.

Self-attention based recurrent convolutional neural network for disease prediction using healthcare data

BACKGROUND AND OBJECTIVE

Nowadays computer-aided disease diagnosis from medical data through deep learning methods has become a wide area of research. Existing works of analyzing clinical text data in the medical domain, which substantiate useful information related to patients with disease in large quantity, benefits early-stage disease diagnosis. However, benefits of analysis not achieved well when the traditional rule-based and classical machine learning methods used; which are unable to handle the unstructured clinical text and only a single method is not able to handle all challenges related to the analysis of the unstructured text, Moreover, the contribution of all words in clinical text is not the same in the prediction of disease. Therefore, there is a need to develop a neural model which solve the above clinical application problems, is an interesting topic which needs to be explored.

METHODS

Thus considering the above problems, first, this paper present self-attention based recurrent convolutional neural network (RCNN) model using real-life clinical text data collected from a hospital in Wuhan, China. This model automatically learns high-level semantic features from clinical text by using bi-direction recurrent connection within convolution. Second, to deal with other clinical text challenges, we combine the ability of RCNN with the self-attention mechanism. Thus, self-attention gets the focus of the model on essential convolve features which have effective meaning in the clinical text by calculating the probability of each convolve feature through softmax.

RESULTS

The proposed model is evaluated on real-life hospital dataset and used measurement metrics as Accuracy and recall. Experiment results exhibit that the proposed model reaches up to accuracy 95.71%, which is better than many existing methods for cerebral infarction disease.

CONCLUSIONS

This article presented the self-attention based RCNN model by combining the RCNN with self-attention mechanism for prediction of cerebral infarction disease. The obtained results show that the presented model better predict the cerebral infarction disease risk compared to many existing methods. The same model can also be used for the prediction of other disease risks.

Early Prediction of Malignant Brain Edema After Ischemic Stroke

Background and Purpose- Malignant brain edema after ischemic stroke has high mortality but limited treatment. Therefore, early prediction is important, and we systematically reviewed predictors and predictive models to identify reliable markers for the development of malignant edema. Methods- We searched Medline and Embase from inception to March 2018 and included studies assessing predictors or predictive models for malignant brain edema after ischemic stroke. Study quality was assessed by a 17-item tool. Odds ratios, mean differences, or standardized mean differences were pooled in random-effects modeling. Predictive models were descriptively analyzed. Results- We included 38 studies (3278 patients) with 24 clinical factors, 7 domains of imaging markers, 13 serum biomarkers, and 4 models. Generally, the included studies were small and showed potential publication bias. Malignant edema was associated with younger age (n=2075; mean difference, -4.42; 95% CI, -6.63 to -2.22), higher admission National Institutes of Health Stroke Scale scores (n=807, median 17-20 versus 5.5-15), and parenchymal hypoattenuation >50% of the middle cerebral artery territory on initial computed tomography (n=420; odds ratio, 5.33; 95% CI, 2.93-9.68). Revascularization (n=1600, odds ratio, 0.37; 95% CI, 0.24-0.57) were associated with a lower risk for malignant edema. Four predictive models all showed an overall C statistic >0.70, with a risk of overfitting. Conclusions- Younger age, higher National Institutes of Health Stroke Scale, and larger parenchymal hypoattenuation on computed tomography are reliable early predictors for malignant edema. Revascularization reduces the risk of malignant edema. Future studies with robust design are needed to explore optimal cutoff age and National Institutes of Health Stroke Scale scores and to validate and improve existing models.

A simple prediction score system for malignant brain edema progression in large hemispheric infarction

Malignant brain edema (MBE) due to hemispheric infarction can result in brain herniation, poor outcomes, and death; outcome may be improved if certain interventions, such as decompressive craniectomy, are performed early. We sought to generate a prediction score to easily identify those patients at high risk for MBE. 121 patients with large hemispheric infarction (LHI) (2011 to 2014) were included. Patients were divided into two groups: those who developed MBE and those who did not. Independent predictors of MBE were identified by logistic regression and a score was developed. Four factors were independently associated with MBE: baseline National Institutes of Health Stroke Scale (NIHSS) score (p = 0.048), Alberta Stroke Program Early Computed Tomography Score (ASPECTS) (p = 0.007), collateral score (CS) (p<0.001) and revascularization failure (p = 0.013). Points were assigned for each factor as follows: NIHSS ≤ 8 (= 0), 9–17 (= 1), ≥ 18 (= 2); ASPECTS≤ 7 (= 1), >8 (= 0); CS<2 (= 1), ≥2 (= 0); revascularization failure (= 1),success (= 0). The MBE Score (MBES) represents the sum of these individual points. Of 26 patients with a MBES of 0 to 1, none developed MBE. All patients with a MBES of 6 developed MBE. Both MBE development and functional outcomes were strongly associated with the MBES (p = 0.007 and 0.002, respectively). The MBE score is a simple reliable tool for the prediction of MBE.

Random Forest

For the task of analyzing survival data to derive risk factors associated with mortality, physicians, researchers, and biostatisticians have typically relied on certain types of regression techniques, most notably the Cox model. With the advent of more widely distributed computing power, methods which require more complex mathematics have become increasingly common. Particularly in this era of "big data" and machine learning, survival analysis has become methodologically broader. This paper aims to explore one technique known as Random Forest. The Random Forest technique is a regression tree technique which uses bootstrap aggregation and randomization of predictors to achieve a high degree of predictive accuracy. The various input parameters of the random forest are explored. Colon cancer data (n = 66,807) from the SEER database is then used to construct both a Cox model and a random forest model to determine how well the models perform on the same data. Both models perform well, achieving a concordance error rate of approximately 18%.