همایش بین المللی هوش مصنوعی

صفحه اصلی / دومین همایش بین المللی هوش مصنوعی

ORB-SVM : An Innovative Hybrid Framework for Efficient Brain Tumor Detection from MRI Scans

نویسندگان :

Amirhosein Azarpour¹

1- دانشگاه شهید بهشتی

کلمات کلیدی :

ORB،Support Vector Machine،Brain Tumor،Data Compression،Medical image processing،MRI images

چکیده :

Brain cancer is among the most formidable types of cancer, posing significant challenges in terms of diagnosis, treat ment, and patient survival.Doctors’ diagnoses are not always reliable, and traditional methods of reading medical images have significant limitations. Recent shifts in medical imaging and machine learning suggest another way forward. MRI scans, already central to visualising the brain’s fine structure, carry a wealth of detail. A single scan can contain tens of thousands of pixels, subtle gradients of grey, overlapping textures that hint at abnormal growth but do not declare it openly. To navigate this complexity, the proposed approach turns to the ORB algorithm, short for Oriented FAST and Rotated BRIEF, as a feature extraction tool. In practical terms, ORB identifies distinctive points in MRI images, such as edges and texture variations that might, say, differentiate a smooth cyst from the ragged outline of a malignant mass. By focusing only on these meaningful points, the method compresses the data drastically, shrinking the original image size by roughly 99.5 percent. The benefit is not just computational speed, though that matters. It also reduces the temptation for the model to latch onto irrelevant noise. Once these refined features are extracted, they are passed into a Support Vector Machine classifier. SVM, with its knack for handling complex and non-linear separations, offers a structured way to distinguish between benign and malignant patterns. The combined ORB-SVM framework, tested here on a substantial MRI dataset, represents an early yet intriguing attempt to bring this pairing into a realistic clinical scenario. The outcomes are, admittedly, impressive. A classification accuracy of 97.5 percent suggests that the model can pick up on differences so slight that they might escape the eye. However, the significant reduction in data size and the resulting efficiency point toward something genuinely useful. Faster analysis, fewer ambiguous interpretations, and the possibility of more timely diagnoses begin to seem not just theoretical, but attainable.