Hierarchical Computer-Aided Detection of Abnormal Regions in Medical Images | Rutgers University Innovation Ventures

Hierarchical Computer-Aided Detection of Abnormal Regions in Medical Images

Receiver operator characteristic curves for detection of prostate cancer at various weights (T) in histological (top) and MRI (bottom) images. Classification is based on multiplicative weighted maximum a posteriori (MWMAP) estimation with iterated conditional modes (ICM). The regions labeled as malignant at three weights are highlighted in green (histology) or red (MRI)


Invention Summary:

Detection of abnormal features in medical images is a time-consuming task performed by pathologists. Although computer analysis can increase throughput, the size of the images often limits processing speeds.

This invention solves these problems using a hierarchical approach. Analysis begins at low resolution to identify abnormal regions and proceeds to higher levels of resolution for these identified regions. Morphological feature information is used to classify these abnormal structures based on probabilistic pairwise Markov models. The classification is enhanced by assigning different weights to each class (ex. normal vs. abnormal), which leads to the selection of regions for high-resolution examination. In this way, the size of regions of interest for high-resolution examination is reduced, thereby reducing processing times.


Market Application:

  • Computer Aided Diagnosis: cancer, pre-malignant lesions, neoplasia, calcification
  • Tissues: prostate, breast, ovary, bladder, kidney, brain
  • Image Modalities: histology, magnetic resonance imaging, mammography, positron emission tomography, computed tomography

Advantages:

  • Rapid processing
  • High throughput
  • Reproducible
  • Quantitative
  • Data mining for biomarker discovery
  • Avoid heuristics

Intellectual Property & Development Status:

Issued US patent. Available for collaboration or licensing.

Patent Information: