BACKGROUND
Classifying brain tumors preoperatively provides essential information for guiding treatment plans. However, existing classification methods require manual intervention which often suffer from efficiency and accuracy issues.
OBJECTIVE
The objective of our study was to Improve the accuracy and efficiency of brain tumor grading or classification. This shows that our method is to help clinicians diagnose brain tumor and formulate treatment plans.
METHODS
Our proposed method uses a DenseNet-ResNet based U-Net framework to optimize the task of extracting features from brain tumor MRI image data. It also adopts a CRNN model to classify brain tumors from sequence data. The characteristic of our method is that it needs only one sequence-level label, instead of many frame-level labels for each patient.
RESULTS
We proposed an automated MRI classification method for brain tumor, with an average accuracy of 90.72% for glioma classification, 94.35% for glioma IDH1 mutation classification, and 94.64% for pituitary tumor texture classification.
CONCLUSIONS
Compared with existing purely auto-encoder based methods, ours has much better efficiency.