Abstract
Stroke remains a leading global cause of mortality, with approximately 5.8 million deaths reported in 2016 alone. Risk factors such as hypertension, diabetes, and age significantly influence the progression of ischemic strokes. This paper presents a hybrid AI framework utilizing Adaptive Neuro-Fuzzy Inference Systems (ANFIS) and Principal Component Analysis (PCA) to predict the optimal timing for decompressive hemicraniectomy. The proposed system achieves high predictive accuracy and perfect precision, offering a robust tool for clinical decision support in emergency neurosurgery.
Introduction
When a patient suffers an ischemic stroke, a blood clot obstructs a cerebral artery, leading to tissue infarction and subsequent edema (brain swelling). Within the rigid confines of the skull, this swelling increases intracranial pressure, which can lead to fatal brain stem compression. Decompressive hemicraniectomy—the surgical removal of a portion of the skull—is a life-saving intervention.
However, its efficacy is highly time-dependent. Surgeons face a dilemma: operating too early may expose a stable patient to unnecessary surgical risks, while operating too late renders the procedure futile. Current monitoring relies on serial CT scans to track infarction growth, a process that consumes valuable time. This research addresses the need for a predictive tool that can determine surgical urgency from minimal initial clinical data.
Methodology
The researchers developed a two-stage hybrid architecture combining dimensionality reduction with fuzzy logic to mimic clinical reasoning. The study utilized a pooled database of 117 stroke patient records from referral centers in Qatar, the UAE, and Pakistan.
- Feature Reduction (PCA): The initial dataset comprised 14 clinical features, including blood pressure, HbA1c levels, and clot burden scores. To mitigate noise and prevent overfitting, Principal Component Analysis (PCA) was employed to distill these into the most significant variables.
- Predictive Modeling: The system utilizes ANFIS, a hybrid model that combines the learning capabilities of neural networks with the interpretability of fuzzy logic. The model predicts the infarction growth rate and the necessity of surgery.
- Fuzzy Inference System (FIS): The FIS acts as the decision-making core. By applying "if-then" rules to predicted growth rates, the system outputs a specific timing recommendation: no surgery, surgery within 48 hours, or surgery within 24 hours.
Key Results and Discussion
The integration of PCA proved vital for model performance. By reducing the complexity of the input space, the ANFIS model achieved significantly lower error rates compared to standard implementations.
The final system demonstrated an overall accuracy of 95.7% in determining surgery timing. Specifically, the decision-making component for surgery necessity reached 97.8% accuracy. Crucially, the model achieved a precision of 1.0 (100%), meaning it produced zero false positives. In a medical context, this ensures that no patient is recommended for high-risk surgery unless it is strictly necessary. The Mean Absolute Error (MAE) for growth rate prediction was reduced to 0.118 with PCA, indicating high reliability in tracking injury progression.
Conclusion
The hybrid ANFIS-PCA approach provides a transparent and highly accurate method for predicting surgical windows in ischemic stroke patients. By combining the data-driven strengths of neural networks with the rule-based logic of clinical expertise, the system offers a viable path toward automated, real-time surgical decision support. Future research will focus on validating these results with larger, multi-center datasets to account for greater physiological diversity.
Summary updated