Purpose

As artificial intelligence continues to be developed and improved upon, it has become an important resource that helps medical professionals diagnose different types of diseases. Our model aids neurologists in identifying brain tumors, which is important for early and accurate detection to ensure proper treatment before the condition worsens. With most tumors, when caught at a low grade, they are usually benign and the intervention is more likely to result in permanent removal of all the cancerous cells. If caught later, at a higher grade, the tumor is more likely to be malignant and is tougher to treat. This highlights the importance of our model, as it allows for quick and accurate classification in order to save time and reduce misdiagnoses.

Dataset

Our dataset contains 3064 images from 233 patients with three types of brain tumors. There are 708 images containing meningioma, 1426 images containing glioma, and 930 images containing pituitary tumors. Each image contains a label indicating the type of tumor (1 for meningioma, 2 for glioma, 3 for pituitary tumor), a patient ID, a vector that contains the coordinates of the tumor border, and a tumor mask, or binary image with 1s indicating the tumor region.

Detection

Our model is able to detect three different kinds of tumors: meningioma, glioma, and pituitary. Each of these are located in the brain, and can have severe effects if not diagnosed early enough. Meningiomas and gliomas are located in the central nervous system, and pituitary tumors are located in the pituitary gland, which affects hormone production.

Results

We found that the overall accuracy of the model was 91%, with a weighted precision of 91%, weighted recall of 91%, and weighted F1-score of 91%. Per class, we can see differences between the model’s ability to classify different types of tumors. For class 1 (meningioma), the precision of the predictions was 84%, with recall being 80%. This suggests that the model has more difficulty correctly identifying these instances. For class 2 (glioma), the model performs well, with 90% precision and 93% recall, indicating strong accuracy in regards to glioma tumors. The model performs best in predicting class 3 (pituitary), with its precision and recall both coming in at 98%.