Improving brain tumor diagnosis: A self-calibrated 1D residual network with random forest integration

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate brain tumor diagnosis relies on precise MRI analysis.
• Existing AI methods for brain tumor identification often lack generalization due to single-dataset dependency.
• There is a need for robust AI tools that perform reliably across diverse clinical data.

Purpose of the Study:

• To introduce SCR-1DResNet, a novel AI diagnostic tool for brain tumor detection.
• To improve the generalization capability and prediction reliability of AI models in brain tumor identification.
• To enhance the speed and diagnostic performance of AI-driven systems for clinical decision support.

Main Methods:

• Utilized MRI image acquisition from multiple Kaggle datasets.
• Implemented stepwise image processing including noise elimination, enhancement, resizing, normalization, and skull stripping.
• Developed the SCR-1DResNet model by integrating self-calibrated Random Forest with one-dimensional residual networks and WaveSegNet for feature extraction.

Main Results:

• Achieved a classification precision of 98.50%.
• Attained an overall accuracy of 98.80%.
• Reached a recall rate of 97.80% for brain tumor detection.

Conclusions:

• The SCR-1DResNet model demonstrates superior diagnostic capability and enhanced performance speed.
• The proposed system shows strong potential for clinical decision support systems.
• SCR-1DResNet offers prospects for improved neurological and oncological patient treatments.