Machine learning for high-energy interactions analysis
View Patent ↗Systems and methods for interpreting high-energy interactions on a sample are described in this application. In particular, this application describes analysis systems and methods, comprising impinging radiation from a source on an analyte, detecting energy interactions resulting from the impinging radiation using a detector, adjusting a signal emitted from the radiation detector using a pre-processing method to emphasize specific features of that signal, using a machine learning module to interpret specific parts of the adjusted signal, producing a quantitative and/or qualitative model using the machine learning module, and applying the quantitative and/or qualitative model to a separate energy interaction. The quantitative and qualitative models derived from this training can be applied to new detector inputs from the same or similar instruments. Other embodiments are described.
1 . An analysis method, comprising:
impinging radiation from a source on an analyte;
detecting energy interactions resulting from the impinging radiation using a detector;
adjusting a signal emitted from the detector using a pre-processing method to emphasize specific or general features of that signal;
using a machine learning module to interpret specific or general parts of the adjusted signal;
producing a quantitative and/or qualitative model using the machine learning module;
applying the quantitative and/or qualitative model to a first set of additional data associated with detected energy interactions resulting from impinging additional radiation to the analyte or a different analyte;
applying a post-processing method to an output from the machine learning module to determine a metric; and
selecting, based on the metric, the pre-processing method, machine learning module, or post-processing method for use in subsequent data processing.
2 . The method of claim 1 , wherein the machine learning module interprets the specific or general parts of the adjusted signal using existing data comprising compositional data or other data characteristics about the analyte through empirical measurements or estimated values or membership of the analyte in a class.
3 . The method of claim 1 , wherein the detector signal is determined by a counting mechanism within the detector that provides data in a continuous interval which can be used either in a full or summarized form by the machine learning module.
4 . The method of claim 1 , wherein the pre-processing method adjusts the detector signal using a normalization method for input mapping that emphasizes features of that signal to produce the adjusted signal, wherein the adjusted signal is provided to the machine learning module which, in turn, interprets the adjusted signal.
5 . The method of claim 1 , wherein the machine learning module uses random forests, gradient boosting, support vector machines, neural networks, or k-nearest neighbors to interpret the adjusted signal.
6 . The method of claim 1 , wherein wherein at least one of the source, the analyte, and the detector has variable geometry.
7 . The method of claim 1 , wherein the radiation from the source includes photons, electrons, or protons with the ability to ionize the analyte.
8 . The method of claim 1 , wherein the first set of additional data associated with detected energy interactions results from impinging additional radiation on the analyte.
9 . The method of claim 1 , wherein the first set of additional data associated with detected energy interactions results from impinging additional radiation on the different analyte.
10 . The method of claim 1 , further comprising using the machine learning module to infer specific parts of the adjusted signal.
11 . The method of claim 1 , wherein adjusting the signal emitted from the radiation detector using the pre-processing method to emphasize specific or general features of that signal comprises using a plurality of pre-processing methods, wherein the plurality of pre-processing methods cooperate to provide a plurality of adjusted signals, wherein selecting, based on the metric, the pre-processing method, machine learning module, or post-processing method for use in subsequent data processing comprises selecting one of the plurality of pre-processing methods, the method further comprising:
obtaining additional data associated with detected energy interactions resulting from impinging additional radiation on the analyte or a different analyte; and
applying the selected pre-processing method to provide a second adjusted signal.
12 . The method of claim 11 , further comprising applying the quantitative and/or qualitative model to the second adjusted signal.
13 . The analysis method of claim 1 , wherein impinging the radiation from the source on the analyte comprises impinging x-rays on the analyte.
14 . The analysis method of claim 1 , wherein impinging the radiation from the source on the analyte comprises impinging gamma rays on the analyte.
15 . An analysis method, comprising:
receiving a signal from a detector that has detected energy interactions from radiation impinging on an analyte;
adjusting the signal emitted from the detector using a pre-processing method to emphasize specific features of that signal;
using a machine learning module to interpret specific parts of the adjusted signal;
producing a quantitative and/or qualitative model using the machine learning module;
applying a post-processing method to an output from the machine learning module to determine a metric; and
selecting, based on the metric, a pre-processing method, machine learning module, or post-processing method for use in subsequent data processing.
16 . The method of claim 15 , further comprising impinging radiation from a source on the analyte.
17 . The method of claim 15 , further comprising applying the quantitative and/or qualitative model to a first set of additional data associated with detected energy interactions resulting from impinging additional radiation on the analyte or a different analyte.
18 . An analysis system, comprising:
a radiation source;
a radiation detector; and
a machine learning module configured to:
receive a signal from the radiation detector that has detected energy interactions from radiation from the radiation source impinging on an analyte;
adjust the signal to emphasize specific parts of the detector signal;
interpret specific parts of the adjusted signal; and
produce a quantitative and/or qualitative model; and
a post-processing module configured to:
determine a metric based on an output from the machine learning module; and
select, based on the metric, a pre-processing method, machine learning module, or post-processing method for use in subsequent data processing.
19 . The system of claim 18 , wherein the machine learning module is further configured to apply the quantitative and/or qualitative model to a separate energy interaction.
20 . The system of claim 18 , wherein the machine learning module is further configured to select a metric that can be used to adjust the signal.
21 . The analysis system of claim 18 , wherein the machine learning module is configured to adjust the signal emitted from the radiation detector using a plurality of pre-processing methods, wherein the plurality of pre-processing methods cooperate to provide a plurality of adjusted signals,
wherein the post-processing module is configured to select, based on the metric, the pre-processing method, machine learning module, or post-processing method for use in subsequent data processing by selecting one of the plurality of pre-processing methods, and
wherein the machine learning module is further configured to:
obtain additional data associated with detected energy interactions resulting from impinging additional radiation on the analyte or a different analyte; and
apply the selected pre-processing method to provide a second adjusted signal.