Abstract
Researchers at the University of Central Florida and the Mayo
Clinic have developed an automated system that enables radiologists to detect
and assess the cancer risk of pancreatic cysts faster and more effectively than
using conventional MRI technology and radiographic guidelines. The cysts, called
intraductal papillary mucinous neoplasms (IPMNs), are radiographically
identifiable precursors to pancreatic cancer. Consequently, the early detection
and risk assessment of IPMNs is vital.
With conventional technology, a radiologist may have to read more than 1,000 images to evaluate a cyst
for just one patient. The new system streamlines the process by using MRI data with machine learning strategies to
improve tumor risk stratification (characterization). Also, the invention can enable non-invasive cancer staging and
prognosis, and foster personalized treatment planning as a part of precision
medicine.
Technical Details
The invention comprises a system and methods for
automatically diagnosing IPMNs in a pancreas
using multi-modal MRI data (TI and T2 images). In one example application, the
system comprises an MRI scanner and a processor programmed to compute the minimum and maximum intensity projections that
correspond to the Tl and T2 MRI scans of a patient's pancreas. The intensity
projections then go into a pre-trained image recognition convolutional neural
network (CNN) algorithm to obtain feature vectors. Next, canonical correlation
analysis (CCA)-based feature fusion is performed
to produce a final vector with discriminative and transformed feature
representation. Finally, the system employs a support vector machine (SVM)
classifier to determine whether the pancreas includes IPMNs.
Benefit
Improves tumor characterization in early detection and precise risk assessment of IPMNsDoes not require manual segmentation of a pancreas or cysts Does not require explicit sample balancingMarket Application
Biomedical imaging and radiologyHospitals and health centers
Brochure
