Yolov8 Model for Brain Tumor Detection Using Combinatorial Multimodal Image Fusion

Abstract

Accurate identification of brain malignancies requires high-precision spatial and spectral data. While multimodal image fusion provides a rich data environment, the optimization of automated detection within these fused datasets remains a critical challenge. This study validates the integration of a fine-tuned YOLOv8s architecture with a Discrete Wavelet Transform (DWT)-based combinatorial multimodal image fusion framework for automated brain malignancy detection. By decoupling the image fusion mechanism from the object detection pipeline, the model effectively leverages enriched spatial and spectral features from combined MRI, CT, and PET modalities. Benchmark evaluations demonstrate that the proposed YOLOv8s significantly outperforms conventional baseline architectures (VGG16, standard CNN, and Faster R-CNN) across all primary evaluation metrics, achieving an mAP50 of 99.77%, an mAP50–95 of 89.83%, a Precision of 98.71%, a Recall of 97.52%, and an F1-score of 94.44%. Compared to single-modality baseline models and models build on non-combination of medical imaging modalities. Experimental results indicate that the integration of YOLOv8 with the established combinatorial fusion framework achieves high diagnostic reliability, successfully minimizing false negatives in complex tumor morphologies. This study validates the efficacy of the proposed YOLOv8 detection model as a robust successor to the initial fusion phase. By decoupling the fusion process from the detection logic, we demonstrate a scalable pipeline for medical imaging diagnostics. The results suggest that this combinatorial approach provides a superior objective evaluation metric for clinical decision support systems in neuro-oncology on the test dataset.