Related Experiment Video
Updated: Oct 2, 2026
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
Parameter Optimized Bi-directional Long Short-term Memory (Pobi-Lstm) for Ich Stroke Detection and Classification on
M Sathya Sundaram1, S Karthick1, P Thiyagarajan2
1Department of Computational Intelligence, School of Computing, Faculty of Engineering and Technology, SRM Institute of Science and Technology, Kattankulathur-603203, Tamil Nadu, India.
Objective:
A life-threatening cerebrovascular disease with high mortality and morbidity is Intracranial Hemorrhage (ICH). Early identification of the different types of ICH through CT Imaging of the Brain is essential for timely treatment. A new method called Parameter Optimized Bidirectional Long Short-Term Memory (POBi-LSTM) has been developed to automatically detect and classify cases of ICH.
Methods:
There are four phases in the proposed framework, which are preprocessing, saliency-based feature mapping, detection/classification, and hyperparameter optimization of the classifier. Initially, CT images will go through adaptive preprocessing before saliency-based features are extracted. Next, a Bi-LSTM classifier will classify the subtypes of Intracranial Hemorrhage (ICH), and Fuzzy Elite Opposition Social Spider Optimization (FEOSSO) will be used to optimize classifier hyperparameters. The proposed framework will be validated using 2500+ Brain CT Images with ICH Masks.
Results:
Experimental evaluation was carried out based on sensitivity, specificity, overall accuracy, Area Under the Curve (AUC), and Matthews Correlation Coefficient (MCC). The Results of the proposed POBi-LSTM framework are as follows: Sensitivity 93.85%, Specificity 94.44%, Accuracy 94.81%, AUC 93.54%, and MCC 93.78%. In comparison to standard techniques including (AlexNet-SVM), Conv-LSTM, 2D-CNNEnsemble, Bi-LSTM, Vision Transformer (ViT), and Swin Transformer, the proposed POBi-LSTM Framework outperformed all these models.
Discussion:
Through the combination of saliency-based feature extraction, bidirectional temporal learning, and hyperparameter optimization using FEOSSO, hemorrhages are effectively characterised in CT images. The high level of performance of the new modeling approach indicates its ability to assist in the accurate and speedy computer-aided diagnosis of intracranial hemorrhages.
Conclusion:
The POBi-LSTM model provides a precise and reliable way of automatically classifying hemorrhage subtype through the use of CT images. This provides radiology with a method that could assist in the early identification and assessment of intracranial bleeding in a clinical setting.