[Paper Review] Convolutional XGBoost (C-XGBOOST) Model for Brain Tumor Detection

Muyiwa Babayomi, Oluwatosin Atinuke Olagbaju+1 more|arXiv (Cornell University)|Jan 5, 2023

Brain Tumor Detection and Classification8 citations

TL;DR

The paper proposes a Convolutional XGBoost (C-XGBOOST) model that combines CNN features with XGBoost for brain tumor detection from MRI, achieving lower model complexity and better handling of imbalanced data than pure CNNs.

ABSTRACT

Brain tumors are masses or abnormal growths of cells within the brain or the central spinal canal with symptoms such as headaches, seizures, weakness or numbness in the arms or legs, changes in personality or behaviour, nausea, vomiting, vision or hearing problems and dizziness. Conventional diagnosis of brain tumour involves some tests and procedure which may include the consideration of medical history, physical examination, imaging tests (such as CT or MRI scans), and biopsy (removal and examination of a small piece of the tumor tissue). These procedures, while effective, are mentally strenuous and time demanding due to the manual examination of the brain scans and the thorough evaluation of test results. It has been established in lots of medical research that brain tumours diagnosed and treated early generally tends to have a better prognosis. Deep learning techniques have evolved over the years and have demonstrated impressive and faster outcomes in the classification of brain tumours in medical imaging, with very little to no human interference. This study proposes a model for the early detection of brain tumours using a combination of convolutional neural networks (CNNs) and extreme gradient boosting (XGBoost). The proposed model, named C-XGBoost has a lower model complexity compared to purely CNNs, making it easier to train and less prone to overfitting. It is also better able to handle imbalanced and unstructured data, which are common issues in real-world medical image classification tasks. To evaluate the effectiveness of the proposed model, we employed a dataset of brain MRI images with and without tumours.

Motivation & Objective

• Motivate early brain tumor diagnosis to improve prognosis.
• Develop a hybrid CNN-XGBoost approach to reduce training complexity.
• Address common dataset issues in medical imaging such as imbalance and unstructured data.

Proposed method

• Integrate convolutional neural networks with Extreme Gradient Boosting to form the C-XGBoost model.
• Aim for lower model complexity and easier training compared to pure CNN approaches.
• Target improved handling of imbalanced and unstructured MRI data in tumor classification.

Experimental results

Research questions

• RQ1 Can the C-XGBoost model effectively detect brain tumors from MRI images?
• RQ2 Does C-XGBoost offer lower model complexity and easier training than purely CNN-based models?
• RQ3 Is C-XGBoost more robust to imbalanced and unstructured data commonly found in medical imaging?

Key findings

• C-XGBoost is proposed as an effective hybrid for brain tumor detection from MRI images.
• The model is described as having lower complexity and easier training than purely CNN-based approaches.
• C-XGBoost is positioned as better suited to handle imbalanced and unstructured medical imaging data.
• The evaluation uses a brain MRI dataset with and without tumors to assess performance.