Diagnosis of disease using laser-induced breakdown spectroscopy and machine learning
Inventors
Guadiuso, Rosalba • Melikechi, Noureddine • Xia, Weiming
Assignees
US Department of Veterans Affairs • University of Massachusetts Amherst
Publication Number
US-12055495-B2
Publication Date
2024-08-06
Expiration Date
2041-07-16
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Abstract
Diagnosing a pathology using LIBS and biological fluids includes focusing light on to a sample on a substrate to cause ablation of the sample and formation of a plasma, collecting optical emission from the plasma, and providing the optical emission to a spectroscopic acquisition component that provides information on spectral data of the plasma. The spectral data is provided to a machine learning algorithm to diagnose a pathology in the sample, where the algorithm is trained on a training set that includes spectral features in a difference spectrum derived from differences between a first LIBS optical emission spectrum collected from one or more samples of the biological fluid that have the pathology or known progress of the pathology and a second LIBS optical emission spectrum collected from one or more samples of the predetermined biological fluid that do not have the pathology or the known progress of the pathology.
Core Innovation
The invention provides systems and methods for diagnosing or monitoring the progress of a pathology using laser induced breakdown spectroscopy (LIBS) and machine learning applied to biological fluids. The method involves depositing a sample of a predetermined biological fluid on a substrate, focusing laser light to ablate the sample and form a plasma, collecting optical emission from the plasma, and providing this spectral data to a processing component employing a machine learning algorithm. This machine learning algorithm is trained on spectral features derived from difference spectra that compare samples with and without the pathology or known progression thereof.
The problem being addressed is the difficulty in early, non-invasive, and cost-effective diagnosis of pathologies such as Alzheimer's disease (AD). AD diagnosis currently relies on costly or invasive procedures like brain imaging or lumbar puncture, or on neurological and cognitive symptom observation, which lacks specificity and early detection capability. There is a need for a minimally invasive, fast, and accurate diagnostic method to detect AD and discriminate it from other dementias early in disease progression.
The use of LIBS combined with machine learning provides a novel approach by analyzing blood plasma or other biological fluids, obtaining spectral data indicative of elemental composition changes associated with the pathology. Difference spectra are used to highlight spectral features distinguishing diseased from healthy samples, enabling training of machine learning algorithms for classification. This approach addresses limitations of previous diagnostic methods by providing a minimally invasive, rapid, and potentially accurate test suitable for early diagnosis and monitoring disease progression.
Claims Coverage
The patent includes two main independent claims covering a method and a system for diagnosing or monitoring a pathology using LIBS and machine learning, with several inventive features specified.
Method for pathology diagnosis using LIBS and machine learning
A method comprising: focusing laser light on a biological fluid sample on a substrate to cause ablation and plasma formation; collecting optical emission from the plasma; providing spectral data to a processing component with processors; and using a machine learning algorithm trained on spectral features from difference spectra comparing pathological and non-pathological samples to screen, diagnose, or monitor pathology. The method includes determining polarity of spectral features relative to a model difference spectrum and classifying samples based on the count of positive versus negative polarity features.
System for pathology diagnosis using LIBS and machine learning
A system comprising: a substrate to support a biological fluid sample; a laser light source focusing on the sample to induce plasma via ablation; an optical subsystem for focusing laser light and collecting plasma emission; a spectroscopic acquisition component with a spectrometer and detector providing spectral data; and a processing component with processors configured to use a machine learning algorithm trained on difference spectra features to screen, diagnose, or monitor pathology. Classification includes determining polarities of spectral features relative to a model difference spectrum and classifying based on positive and negative polarity features.
Use of blood or blood plasma and Alzheimer's disease pathology
The biological fluid can include blood or blood plasma, and the pathology diagnosed can be Alzheimer's disease.
Application of Quadratic Discriminant Analysis algorithm
The machine learning algorithm used can include a Quadratic Discriminant Analysis (QDA) algorithm.
Weighting spectral features based on classification ability
Spectral features used in training the machine learning algorithm are weighted according to their ability to correctly classify samples.
Labelling spectral features by polarity
Each spectral feature is labelled as indicating or contra-indicating the pathology or progression based on its polarity in the difference spectrum.
Normalization of spectra for total emission intensities
The first and second LIBS optical emission spectra used to generate difference spectra are normalized for their total emission intensities.
Selection of spectral features relative to substrate spectrum
Spectral features used exclude those present in the substrate spectrum or include only features absent in the substrate spectrum or with intensity less than 50% of average intensity in pathological samples.
Generation of difference spectra to enhance spectral differences
Spectral data provided to the processing component includes generating difference spectra to enhance differences between samples from patients with pathology and healthy controls.
The independent claims cover both a method and a system employing LIBS and machine learning for diagnosis or monitoring of pathology using spectral data and difference spectra. They include specific inventive features such as use of difference spectra, feature weighting, polarity-based labelling, normalization, and feature selection relative to substrate spectra, with application especially to blood plasma and Alzheimer's disease, and employing algorithms like QDA for classification.
Stated Advantages
Provides an automated procedure with speed of analysis and high classification accuracy.
Offers a minimally invasive and cost-effective approach using easily harvested biological fluids.
Enables early diagnosis and monitoring of disease progression, in contrast to costly or invasive prior methods.
Machine learning enhances discrimination capability based on spectral features from plasma emission.
The approach supports unbiased survey of spectroscopic signatures across multiple biological markers.
Documented Applications
Diagnosis and monitoring of Alzheimer's disease using blood plasma.
Diagnosis of Gulf War Illness using blood plasma samples of veterans.
Interested in licensing this patent?