Comparison of MRS analysis using three different dataset reduction schemes. The white box corresponds to the potential cancer space defined from whole-mount histological sections. Colors correspond to the classification of tissue: Red = cancerous, blue = benign, green = indeterminate. Left: z-score; Middle: Principal component analysis (PCA); Right: Nonlinear dimensionality reduction (NLDR).
Combining structural information from Magnetic Resonance Imaging (MRI) with metabolic information from Resonance Spectroscopy (MRS) can greatly improve detection of cancer. However, it is difficult to extract meaningful clinical information from this extensive dataset in an automated, reproducible manner.
Researchers at Rutgers have developed a method to integrate MRI and MRS data to identify regions of cancerous tissue. The structure of interest is first segmented using a hierarchical clustering algorithm. Next, a dataset including MRS spectra and MRI image features corresponding to this structure is extracted. The dataset is then reduced using nonlinear dimensionality reduction (NLDR) to classify cancerous and healthy tissue. This approach can rapidly detect prostate cancer from combined MRI and MRS data.
- Non-invasive screening
- Computer-aided diagnosis
- Image-guided therapy
- Improved sensitivity and specificity
- Rapid (<20s per image)
- Fully automated and reproducible
- Accurate segmentation of the prostate gland
Intellectual Property & Development Status:
Issued US patent. Available for licensing and/or research collaboration.
Tiwari, P. et. al., (2009). Medical Physics, 35:3927-39.