Author: William F.B Igoniye, Belema Manuel, Christopher F. Njeh, O Ray-offor 👨🔬
Affiliation: Indiana University School of Medicine, Department of Radiation Oncology, Department of Radiology, University of Port Harcourt Teaching Hospital 🌍
Purpose: The accurate and efficient categorization of brain tumors is essential for effective treatment planning and improved patient outcomes. Current MRI-based diagnostic methods are time-intensive and subjective to interobserver variability. This study aims to develop a deep learning (DL) model for the automated classification of brain tumours, compare its performance against traditional Radiomics approaches and human expert interpretation , and validate its potentail for integration into clinical practice and to advance precision medicine in neuroimaging.
Methods: A large dataset of brain MRIscans with histological confirmation was collected and preprocessed. A convolutional neural network (CNN)- based DL-model was designed and trained to classify brain tumors into gliomas, meningiomas, and metastases . Hyperparameter optimization and regularization techniques were employed to maximize performance. Model validation was performedon an independent test set , with metrics including accuracy, sensitivity, and specificity calculated. The model's performance was benchmarked against traditional radiomics - based classification methods and expert radiologist interpretations.
Results: The DL model demonsrtrated high accuracy in classifying brain tumors: giomas (accuracy:92%,sensitivity: 91% Specificity :93%), meningiomas ( accuracy 87%, sensitivity:85%, specificity:89%), and metastases ( accuracy: 90%, sensitivity :88%, specificty :92%). Comparative analysis revealed the DL model outperformed radiomics approaches and closely matched expert-level diagnostic accuracy. The robustnessof themodelwas confirmed through repeated cross -validation, indicating its potential for generalizability across diverse patient populations
Conclusion: This study highlights the efficacy of deep learning in automating thecategorisation ofbrain tumors using MRI, offering a scalable solution to improve diagnostic accuracy and efficiency . By reducing interobserver variability and accelerating the diagnoostic process,the proposed DL model can signficantly enhance treatments planning and monitoring inneuro-oncology, paving the way for its adoption in clinical practice. These advancement underscore the potential of DL in precision medicine , promising betteroutcomes for patients with brain tumors.