قطعه‌بندی معنایی تومورهای مغزی از تصاویر ام.آر.آی

Brain tumor is one of the biggest brain diseases known to human. Being able to detect the cancer in its early stages is crucial for patient’s survival. Several methods have been introduced in order to detect the brain tumor, and the most effective one so far, is using MRI images as they clearly indicate the presence of tumor and its parts. However, as the detectors are humans, there is place for error and misunderstanding of the tumor. Hence, just like solving other problems, scientists have been using AI to detect and segment brain tumor parts.

As the rise of convolutional neural networks, they have shown great accuracy in various datasets. The BraTS (Brain Tumor Segmentation) dataset however, has very little amounts of data. Thus, we face overfitting in our model as there is very little valuable information in the third dimension when segmenting (the MRI images have 3 dimensions). The other issue we face, is the high amount of memory usage of the 3D CNN model we use and the problem here is that most people’s GRAM is limited due to its high cost.

In this project, we try to solve both of these problems by changing the basic 3D UNet architecture to confront the problems.

اعضای پروژه:

• محمدرضا رضائی
• نیلوفر آذری
• احسان قره‌آغاجی
• فرزاد حلاجی آزاد
• امیرحسین اقدم