Medical object detection is one of the important methods used in the detection and diagnosis of diseases. In this study, a solution was sought for a bias problem that was not noticed in the testing environment from object detection-based brain MRI studies but appeared in real life. While in the classical data processing pipeline, deep learning model training and testing are performed with only tumor data, a new pipeline in which both tumor and brain parts are labeled is proposed in this study. A state-of-the-art object detection model Mask RCNN was chosen as the deep learning model. According to the results obtained using the BraTS 2020 dataset, in the classical data processing method, specificity value and F1 Score were 0.60 and 0.80, respectively, while the proposed approach measured as 0.94 for both. So the proposed pipeline plays an essential role in reducing false positives, which we frequently encounter in real-life implementations.