Advanced finite segmentation model with hybrid classifier learning for high-precision brain tumor delineation in PET
K Murugan1, SatheeshKumar Palanisamy2, N Sathishkumar3
1Department of ECE, KPR Institute of Engineering and Technology, Coimbatore, 641407, India.
Scientific Reports
|July 15, 2025
Summary
A new Finite Segmentation Model with Improved Classifier Learning (FSM-ICL) significantly boosts brain tumor segmentation accuracy in PET images. This AI approach achieves 92.57% accuracy, outperforming current methods for better diagnostics and treatment planning.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computational Biology
Background:
- Accurate brain tumor segmentation is vital for clinical diagnostics and treatment planning.
- Challenges include complex tumor structures and variations in imaging modalities, hindering precise tumor detection.
- Existing methods struggle with segmentation discreteness and feature differentiation across varying extraction rates.
Purpose of the Study:
- To enhance segmentation accuracy in Positron Emission Tomography (PET) images using a novel Finite Segmentation Model (FSM) with Improved Classifier Learning (ICL).
- To improve the precision of tumor detection by addressing segmentation discreteness and enhancing feature differentiation.
- To provide a more accurate and efficient tool for automated tumor detection and AI-driven medical imaging.
Main Methods:
- Developed the Finite Segmentation Model (FSM) integrated with Improved Classifier Learning (ICL).
- Employed advanced textural feature extraction and deep learning-based classification within an adaptive segmentation framework.
- Trained and validated the FSM-ICL model on the Synthetic Whole-Head Brain Tumor Segmentation Dataset (1000 training, 426 testing images).
Main Results:
- Achieved a segmentation accuracy of 92.57%, significantly outperforming existing methods (NRAN: 62.16%, DSSE-V-Net: 71.47%, DenseUNet+: 83.93%).
- Enhanced classification precision to 95.59% with a reduced classification error of 5.67%.
- Demonstrated a 10.09% improvement in precision and a 10.96% boost in classification rates over state-of-the-art techniques, with a classification time of 572.39 ms.
Conclusions:
- The FSM-ICL framework offers superior performance in brain tumor segmentation from PET images, effectively addressing segmentation discreteness.
- The hybrid classifier learning approach ensures continuous and discrete tumor region detection with enhanced feature differentiation.
- This advancement holds significant potential for automated tumor detection, personalized treatment strategies, and AI-driven medical imaging innovation.
