Meta-learning based Diagnosis of Rare Cancers from MRI Images
 
The prospective summer intern will work on the predictions of rare cancers from radiology images. There are over 100 types of cancers that affect humans. Most of the cancers cause tumors due to abnormal cell division. These tumors can be visualized using different modalities of radiology imaging such as Positron Emission Tomography (PET), Magnetic Resonance Imaging (MRI), and Computed Tomography (CT). Radiologists can make predictions about the presence of cancer by analyzing those images. However, there are several types of cancers that are rarely found in the human body. Therefore, radiology images of the tumors, associated with such cancers are scarcely available. Due to their unfamiliarity, making predictions about such rare cancers poses a significant challenge for the radiologists in clinical practice. The goal of this project is to design a meta-learning framework where an artificially intelligent (AI) system will make predictions about such rare cancers from MRI images with the help of the knowledge, learned form only a few or no images of those cancers. The AI system will make use of some auxiliary information about different cancers to aid the above decision-making process. Under the supervision of a post-doctoral fellow, the intern is expected to conceive and implement the different modules of the above AI system; and work with the clinicians to evaluate its accuracy. At the end of the internship, the intern will be able to learn state-of-the-art machine learning techniques and implement them to solve a challenging real-world problem.
 
The intern will be supervised by my postdoctoral fellow Angshuman Paul, Ph.D.