SPBTGNS: Design of an Efficient Model for Survival Prediction in Brain Tumour Patients Using Generative Adversarial Network With Neural Architectural Search Operations
Open Access
- 17 July 2024
- journal article
- research article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 140847-140869
Abstract
The landscape of medical imaging, particularly in brain tumor analysis and survival prediction, necessitates advancements due to the inherent complexities and life-threatening nature of brain tumors. Existing methodologies often struggle with precision and efficiency, predominantly due to limitations in handling the diverse and intricate nature of medical image datasets. This research presents a novel approach that aims to improve the accuracy of survival prediction in patients with brain tumours, leveraging a Generative Adversarial Network (GAN) integrated with Neural Architectural Search (NAS) operations. The model employs the Adaptive Computation Time (ACT) Transformer, a method crucial for dynamically adjusting the number of transformer layers based on the complexity of the input sets. This feature is particularly beneficial in medical imaging for adapting to the varying complexities of brain tumor data samples. The integration of Squeeze-and-Excitation Networks (SENet) enables the model to recalibrate features channel-wise, significantly enhancing sensitivity to pivotal features in MRI images. Furthermore, the application of Google’s AutoML Vision Edge offers efficient neural architecture and hyperparameter optimization, specifically tuned for image data samples. Efficient Neural Architecture Search (ENAS) is utilized to discover high-performance models with lower computational demands, a critical aspect for medical image analysis where resource constraints are common for different use cases. The model also incorporates customized loss functions, particularly Weighted Cross-Entropy Loss, addressing the class imbalance in medical datasets by emphasizing rarer tumor types. Spatial Dropout and Batch Normalization are integrated as regularization techniques to improve generalization and reduce overfitting risks. The model’s efficacy was validated on the Br35H, Kaggle Brain Tumor Dataset, and IEEE Data Port Dataset Databases, exhibiting a notable improvement over existing methods: 5.9% better precision, 6.5% higher accuracy, and 4.9% higher recall in brain tumor analysis. In survival analysis, the model demonstrated 8.5% better precision and 8.3% higher accuracy, among other improvements. These enhancements underscore the model’s capability in providing more accurate, efficient, and reliable predictions for brain tumor patients, potentially revolutionizing the approach to brain tumor diagnosis and survival prognostication in clinical settings.Keywords
This publication has 42 references indexed in Scilit:
- Brain Tumor Detection and Multi-Grade Segmentation Through Hybrid Caps-VGGNet ModelIEEE Access, 2023
- Brain Tumor Classification Using Hybrid Single Image Super-Resolution Technique With ResNext101_32× 8d and VGG19 Pre-Trained ModelsIEEE Access, 2023
- Prediction of Survival of Glioblastoma Patients Using Local Spatial Relationships and Global Structure Awareness in FLAIR MRI Brain ImagesIEEE Access, 2023
- Brain Tumor Detection and Classification Using Intelligence Techniques: An OverviewIEEE Access, 2023
- Multiple Types of Cancer Classification Using CT/MRI Images Based on Learning Without Forgetting Powered Deep Learning ModelsIEEE Access, 2023
- Salvage craniospinal irradiation for recurrent intracranial germinoma: a single institution analysisJournal of Radiation Research, 2023
- Automated Segmentation of Brain Tumor MRI Images Using Deep LearningIEEE Access, 2023
- A Robust Approach for Brain Tumor Detection in Magnetic Resonance Images Using Finetuned EfficientNetIEEE Access, 2022
- VGG-SCNet: A VGG Net-Based Deep Learning Framework for Brain Tumor Detection on MRI ImagesIEEE Access, 2021
- Brain extraction on MRI scans in presence of diffuse glioma: Multi-institutional performance evaluation of deep learning methods and robust modality-agnostic trainingNeuroImage, 2020