PMRAM Brain Tumor Detection Dataset (Annotated)
Description
This dataset provides bounding box annotations for object detection of brain tumors using MRI scans. It is a derivative work based on the original "PMRAM: Bangladeshi Brain Cancer - MRI Dataset". The annotations were created to facilitate the development, training, and evaluation of machine learning and deep learning models for automated brain tumor detection. Dataset Contents: - Total Images: 1,505 - Total Annotations (Bounding Boxes): 1,074 - Format: COCO JSON (instances_default.json) - Classes: Glioma (344 boxes), Meningioma (359 boxes), Pituitary (371 boxes), Normal (0 boxes - included as negative samples). Annotation Methodology: We manually drew bounding boxes around the visible tumor boundaries. The dataset includes "Normal" images containing no bounding boxes (negative samples) to prevent model hallucination and improve specificity. Original Source: The underlying MRI images are from the "PMRAM: Bangladeshi Brain Cancer - MRI Dataset" created by Md Shahriar Mannan, Prot toy; Chowdhury, Mahtab; Rahman, Redwan; Tamim, Azim Ullah; Rahman, Md Mizanur (2024). Original Dataset DOI: 10.17632/m7w55sw88b.1
Files
Steps to reproduce
1. Download and unzip the dataset. 2. Load the instances_default.json file using a COCO-compatible deep learning framework (e.g., Detectron2, MMDetection, or Ultralytics YOLO). 3. Ensure the images folder is in the same relative directory as the JSON file.
Institutions
- Yunnan UniversityYunnan, Kunming