Lightweight ResNet-18 Based Brain Tumor Classification on MRI with Mixed Precision and Augmentation

Le résumé fourni par la source

In medical imaging, brain tumor identification is a crucial task that has a big influence on early diagnosis and treatment results. A common component of traditional diagnostic techniques is the manual interpretation of CT or MRI scans, which is laborious and prone to human error. Using a refined ResNet-18 convolutional neural network, this study provides a deep learning-based method for automated brain tumor classification in this paper. To enhance generalization and performance, the study used sophisticated training techniques such as label smoothing, cosine annealing learning rate scheduling, mixed precision training, and significant data augmentation, all of which capitalize on transfer learning from ImageNet. A carefully selected collection of brain tumor pictures is used to train and assess the proposed model, which yielded a top accuracy of 98.12%. The model’s robustness is further demonstrated by statistical analysis that includes a confusion matrix, classification report, precision-recall curve, and inference time assessment. This study demonstrates that even very lightweight designs, such as ResNet-18, may produce high diagnostic accuracy when properly tuned, making them appropriate for use in clinical settings with limited resources.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.