Machine-learning algorithm to improve disease diagnosis using magnetic resonance spectroscopy

This technology is a neural network architecture that quantifies in vivo metabolites from magnetic resonance spectroscopy for improved disease detection and diagnosis.

Unmet Need: Accurate and automated method to measure metabolites in clinical imaging

Magnetic resonance spectroscopy (MRS) enables non-invasive quantification of tissue metabolites from magnetic resonance imaging (MRI) scans. When integrated through advanced clinical imaging platforms, these modalities enable the identification of disease-associated metabolites. Thus, this technique has been increasingly used to diagnose cancers, neurodegenerative disorders, and infectious diseases. However, current automated methods for metabolite detection and quantification often suffer from limited accuracy due to poor image quality. Further, manual analysis is time-consuming and impractical for routine clinical use.

The Technology: Deep learning algorithm for improved metabolite quantification compatible with MRI

This technology describes a deep learning algorithm capable of automatically quantifying metabolites from magnetic resonance spectroscopy (MRS) data. The algorithm can be efficiently trained using either in vivo or simulated datasets, enabling rapid, high-throughput, and accurate analysis. It integrates with existing clinical imaging platforms and has the potential to enhance both diagnostic and prognostic capabilities. The algorithm eliminates reliance on time-consuming and error-prone human data interpretation and outperforms existing methods by accurately analyzing low-quality or noisy MRS data.

This technology has been validated with human MRI data.

Applications:

• Tumor detection and diagnosis
• Improved prediction and diagnosis of prostate cancer
• Diagnosis for neurological and neurodegenerative diseases such as Alzheimer’s disease, dementia, Parkinson’s disease, multiple sclerosis, and epilepsy
• Research tool for studying the role of metabolites in disease
• Clinical evaluation tool to monitor cancer treatment effectiveness
• Tool for early stroke detection
• Diagnosis of infectious diseases, such as HIV associated encephalopathy

Advantages:

• Automated, unbiased algorithm that does not require human intervention for metabolite analysis
• High throughput for large datasets
• Faster data processing, resulting in a potential earlier diagnosis
• Simultaneous multi-metabolite identification
• Potential for integration with other existing imaging techniques
• Improved ability to interpret noisy or imperfect datasets
• Flexibility to be trained in various ways depending on data availability
• Compatible with current clinical protocols

Lead Inventor:

Scott A. Small, M.D.

Patent Information:

Patent Pending (US20220051801)

Related Publications:

• Ma DJ, Yang Y, Harguindeguy N, Tian Y, Small SA, Liu F, Rothman DL, Guo J. “Magnetic Resonance Spectroscopy Spectral Registration Using Deep Learning.” J Magn Reson Imaging. 2024 Mar;59(3):964-975.

Tech Ventures Reference:

• IR CU24256, CU24266

• Licensing Contact: Jerry Kokoshka