The primary dataset is sourced from the Kaggle "Brain Tumor MRI Dataset" by Masoud Nickparvar.
Additional "unknown" category samples were collected from publicly available online sources to evaluate robustness.

Training

20 epochs using Hugging Face Trainer
Learning rate: 3e-5
Augmentation: flips, rotation, brightness/contrast jittering
Optimizer: AdamW
Evaluation metric: Accuracy

Performance

Class	Precision	Recall	F1-score
Glioma	0.97	0.98	0.98
Meningioma	0.97	0.96	0.97
No Tumor	0.99	1.00	0.99
Pituitary	0.99	0.99	0.99
Unknown	1.00	0.99	1.00

Overall accuracy: 98 percent

How to Use

python

from transformers import AutoModelForImageClassification, AutoImageProcessor
from PIL import Image
import torch

model = AutoModelForImageClassification.from_pretrained("itistamtran/vit_brain_tumor_multiclass_v2")
processor = AutoImageProcessor.from_pretrained("itistamtran/vit_brain_tumor_multiclass_v2")

image = Image.open("mri_image.png")
inputs = processor(images=image, return_tensors="pt")

with torch.no_grad():
    outputs = model(**inputs)
    preds = outputs.logits.argmax(dim=1)
print(preds)