Enhanced Medical Anomaly Detection Using Particle Swarm Optimization-based Hybrid MLP-LSTM Model
Dharma Teja Valivarthi, Sai Sathish Kethu, Durai Rajesh Natarajan, Swapna Narla, Sreekar Peddi, Aravindhan Kurunthachalam · International Journal of Pattern Recognition and Artificial Intelligence · 2025
The early diagnosis and treatment proceedings can be taken care of with medical anomalies and particularly with brain tumor identification. This work proposes the two-wave tsunami-difference hybrid MLP (Multi-Layer Perceptron) model coupled with PSO (Particle Swarm Optimization). “Two-wave tsunami-difference” is a new feature extraction method that uses two different wave-based transformations to extract difficult-to-see patterns in MRI (Magnetic Resonance Imaging) images for better detection. Improved accuracy and robustness in medical anomalies detection are its basic features. With robust spatial feature extraction, the MLP and LSTM (Long Short-Term Memory) build sequential dependencies in medical imaging data. PSO hyperparameter optimizes the convergence of the model by minimizing false positives and negatives. Additionally, computational complexity is reduced. The model is trained on MRI scans having brain tumors classified into Glioma, Pituitary, and Normal categories. More than the methods available in the literature, the efficacy of the proposed method was shown in terms of accuracy at 99.45%, recall at 99.51%, and precision at 99.41%. Moreover, it reflected 99.46% in terms of F1 score and achieved 98.93% AUC-ROC. Comparative analysis is being conducted considering additional cutting-edge models like LR-HGBC-CNN, ADASYN-TL, and VGG16 for determining the efficacy of this approach in improving the classification accuracy while reducing the misclassification rate. Performance evaluation using ROC curves, Precision-Recall curves, and confusion matrices confirms the robustness of the model. This proposed framework is also a scalable, interpretable, and computationally efficient model for medical anomaly detection, for clinical decision-making purposes.