Exécution de notre outil de synthèse sur un nouvel article

Vérifiez l’article ci-dessous, puis lancez sa synthèse.


Comparative Analysis of MobileNet Architectures for Brain Tumor Detection

Beena Suresh Gaikwad, ALLI Praise A., Vasantha Kumari N

International Conference on Advances in Next-Gen Computer Science (ICANCS)
Licence : non indiquée (tous droits réservés)

Résumé : Early and correct diagnosis of brain tumors is crucial for improved treatment outcomes and patient survival. Early diagnosis, however, is a major challenge in resource-constrained clinical environments. Although deep learning has shown significant promise in the medical image classification task, the most effective lightweight model to apply for real-time processing is still an open issue. In this paper, we investigate which version of MobileNetV1, MobileNetV2, MobileNetV3-Small and MobileNetV3-Large is best suited for efficient and accurate brain tumor classification from contrast-enhanced T1-weighted MRI scans. Each model is fine-tuned with transfer learning utilizing weights pretrained on ImageNet, where they are applied on a dataset of$\mathbf{7, 0 2 2 ~ M R I}$scans. We perform testing and comparison of the models on different performance metrics such as accuracy, precision, recall,$F 1$-score, training time, and model size. It is shown that MobileNetV2 offers the optimal trade-off with training accuracy of 99.60 %, validation accuracy of 99.43 %, and test accuracy of 99.17 % utilizing a small model size and low computational expense. The findings offer important insights on the practical application of deep learning models in clinical environments where computational efficiency is of utmost priority. Future studies will extend the study to other ensemble models and investigate the use of multi-modal MRI sequences and explainable AI to improve diagnostic robustness and accuracy.

Publié le 14 Nov. 2025