Early and accurate detection of brain tumors is crucial for improving clinical outcomes. However, classical MRI-based diagnosis suffers from noise, low contrast, and weak tumor features, posing a challenge to manual and conventional automatic methods. Here, these challenges are addressed by taking into consideration the performance of the YOLOv8 deep learning algorithm, which has been enhanced with strict image preprocessing and data augmentation. With a labeled brain MRI data set, we evaluated four model configurations: (1) raw input, (2) preprocessing only, (3) preprocessing then augmentation, and (4) augmentation then preprocessing. The techniques employed are Gaussian and median blurring, adaptive histogram equalization, morphological operations, and the watershed algorithm, as well as geometric and photometric augmentations. The configuration involving preprocessing followed by augmentation achieved the best performance for YOLOv8 at 91.9% precision, 87.9% recall, and 93.2% mAP50. Subsequently, YOLOv11 on the same enhanced dataset achieved 93.3% precision, 88.2% recall, and 93.6% mAP50. The above results evidently demonstrate that meticulous image enhancement significantly improves deep learning-based tumor detection, with a robust pipeline for medical imaging tasks.