Lists
Lists
Hussain, Tarak, Virdee, Bal Singh, Gudapati, Divya, Ansari, Md. Shamsul Haque and Lakkakula, Sailakshmi (2025) BPBO-LSTM-BiGRU: generative adversial network with brood parasitism-based optimization for spinal muscular atrophy using multiple visual modalities. International Journal of Information Technology. pp. 1-8. ISSN 2511-2112
Spinal Muscular Atrophy (SMA) disorder is characterized by progressive muscle weakness due to motor neuron loss in the spinal cord and brainstem. The studies finds that more accurate diagnosis and classification of SMA require the integration of multiple imaging modalities, such as brain MRI and spinal CT scans, to capture comprehensive structural abnormalities. In this context, a fusion of data augmentation, feature selection along with robust multimodal deep learning (DL) architecture is presented to analyses visual inputs to increase the detection accuracy of spinal muscular atrophy. In this study, we proposed a novel hybrid framework that leverages Generative Adversarial Networks (GANs) for data augmentation, Brood Parasitism-Based Optimization (BPBO) for feature selection, and a hybrid LSTM-BiGRU (long short term memory- gated recurrent unit) model for robust classification. The stacked model is trained on extracted features from multi-modal images, effectively distinguishing between different SMA severity levels. The proposed methodology is rigorously evaluated using a dataset of brain MRI and spinal CT scans, achieving superior classification performance compared to traditional approaches. The performance report demonstrates that the GAN-enhanced dataset with improved feature space enhances classification accuracy of the LSTM-BiGRU network. Various performance metrics, including accuracy, Cohen kappa score and Jaccard Score, reported in the result section indicates that the proposed method outperforms conventional feature selection and classification techniques.
![[thumbnail of Accepted-version.pdf]](https://repository.londonmet.ac.uk/style/images/fileicons/text.png "Accepted-version.pdf")
Restricted to Repository staff only until 9 August 2026.
Download (2MB) | Request a copy
 |
View Item |