Multispectral detection and classification of bacteria utilizing scattering and/or absorbance, excitation, emissions utilizing machine learning
The example techniques and mechanisms described herein can enable relatively inexpensive and/or relatively fast classification of aqueous samples of bacteria or viruses. The example techniques and mechanisms described herein can enhance prior generations of miniature spectrophotometers to include Mie scattering effects. The example techniques and mechanisms described herein can provide a relatively small increase in cost over prior approaches while providing improved classification accuracy based on addition of Mie scattering data to absorbance and emissions data.
1 . A method for classification of an aqueous sample having at least one bacteria species, the method comprising:
applying a selected subset of available wavelengths to the aqueous sample;
measuring an excitation-emission matrix (EEM) corresponding to the selected wavelengths applied to the aqueous sample;
extracting a scattering (D) spectrum from the excitation-emission matrix, wherein the scattering spectrum comprises at least a Rayleigh scattering spectra or at least a Mie scattering spectra;
extracting one or more raw features from the excitation-emission matrix and the scattering spectrum for the aqueous sample;
performing machine learning operations corresponding to the aqueous sample based on the extracted raw features;
generating classification information for bacteria species of the aqueous sample based on results of the machine learning operations and one or more of the excitation-emission matrix and the scattering spectrum of the aqueous sample; and
generating an output indicating the classification information for the bacteria species in the aqueous sample.
2 . The method for classification of claim 1 further comprising:
measuring an absorbance (A) corresponding to the selected wavelengths applied to the aqueous sample;
extracting one or more raw features from the absorbance measured for the aqueous sample;
performing machine learning operations corresponding to the aqueous sample based on the extracted raw features;
generating classification information for bacteria species of the aqueous sample based on results of the machine learning operations and one or more of the excitation-emission matrix and the scattering spectrum of the aqueous sample; and
generating an output indicating the classification information for the bacteria species in the aqueous sample.
3 . The method for classification of claim 1 further comprising:
extracting at least one of an excitation (X) spectrum or an emissions (E) spectrum from the EEM;
extracting one or more raw features from the excitation spectrum and/or the emissions spectrum measured for the aqueous sample;
performing machine learning operations corresponding to the aqueous sample based on the extracted raw features;
generating classification information for bacteria species of the aqueous sample based on results of the machine learning operations and one or more of the excitation-emission matrix and the scattering spectrum of the aqueous sample; and
generating an output indicating the classification information for the bacteria species in the aqueous sample.
4 . The method for classification of claim 1 wherein the machine learning operations are based on intensity versus wavelength at a fixed angle.
5 . The method for classification of claim 1 wherein the classification is based on intensity versus wavelength at a fixed angle.
6 . The method for classification of claim 1 wherein the selected subset of wavelengths comprise 260 nm, 300 nm, 370 nm, 400 nm, 500 nm, and 600 nm.
7 . The method for classification of claim 1 wherein scattering measurements are supplemented with estimates of a spectra derivative by sampling incident and scattered light at λ exc −Δλ, λ exc and λ exc +Δλ, where λ exc is the nominal excitation wavelength and Δλ is the spectrometer resolution.
8 . A spectrometer system comprising:
a light source array to project light at selected wavelengths;
a sample platform to hold an aqueous sample having bacteria or viruses to be classified;
a spectrometer to receive light from the light source array along a direct optical path with respect to the aqueous sample and to receive light from the light source along a 90-degree optical path with respect to the sample; and
a compute engine coupled with the spectrometer, the spectrometer and the compute engine together to:
apply a selected subset of available wavelengths to the aqueous sample;
measure an excitation-emission matrix (EEM) corresponding to the selected wavelengths applied to the aqueous sample;
extract a scattering (D) spectrum from the excitation-emission matrix, wherein the scattering spectrum comprises at least a Rayleigh scattering spectra or at least a Mie scattering spectra;
extract one or more raw features from the excitation-emission matrix and the scattering spectrum for the aqueous sample;
perform machine learning operations corresponding to the aqueous sample based on the extracted raw features;
generate classification information for bacteria species of the aqueous sample based on results of the machine learning operations and one or more of the excitation-emission matrix and the scattering spectrum of the aqueous sample; and
generate an output indicating the classification information for the bacteria species in the aqueous sample.
9 . The spectrometer system of claim 8 wherein the compute engine and the spectrometer are further configured to:
measure an absorbance (A) corresponding to the selected wavelengths applied to the aqueous sample;
extract one or more raw features from the absorbance measured for the aqueous sample;
perform machine learning operations corresponding to the aqueous sample based on the extracted raw features;
generate classification information for bacteria species of the aqueous sample based on results of the machine learning operations and one or more of the excitation-emission matrix and the scattering spectrum of the aqueous sample; and
generate an output indicating the classification information for the bacteria species in the aqueous sample.
10 . The spectrometer system of claim 8 wherein the compute engine and the spectrometer are further configured to:
extract at least one of an excitation (X) spectrum or an emissions (E) spectrum from the EEM;
extract one or more raw features from the excitation spectrum and/or the emissions spectrum measured for the aqueous sample;
perform machine learning operations corresponding to the aqueous sample based on the extracted raw features;
generate classification information for bacteria species of the aqueous sample based on results of the machine learning operations and one or more of the excitation-emission matrix and the scattering spectrum of the aqueous sample; and
generate an output indicating the classification information for the bacteria species in the aqueous sample.
11 . The spectrometer system of claim 8 wherein the machine learning operations are based on intensity versus wavelength at a fixed angle.
12 . The spectrometer system of claim 8 wherein the classification is based on intensity versus wavelength at a fixed angle.
13 . The spectrometer system of claim 8 wherein the selected subset of available wavelengths comprise wavelengths between 250 nm and 600 nm.
14 . The spectrometer system of claim 13 wherein the selected subset of available wavelengths between 250 nm and 600 nm comprise 260 nm, 300 nm, 370 nm, 400 nm, 500 nm, and 600 nm.
15 . The spectrometer system of claim 8 wherein scattering measurements are supplemented with estimates of a spectra derivative by sampling incident and scattered light at λ exc −Δλ, λ exc and λ exc +Δλ, where λ exc is the nominal excitation wavelength and Δλ is the spectrometer resolution.
16 . A compute engine comprising:
a spectrometer interface; and
processing logic coupled with the spectrometer interface, the processing logic to receive via the spectrometer interface spectrometer data comprising at least an excitation-emission matrix (EEM) corresponding to a selected subset of available wavelengths for an aqueous sample having bacteria or viruses to be classified, the processing logic to:
extract a scattering (D) spectrum from the excitation-emission matrix, wherein the scattering spectrum comprises at least a Rayleigh scattering spectra or at least a Mie scattering spectra;
extract one or more raw features from the excitation-emission matrix and the scattering spectrum for the aqueous sample;
perform machine learning operations corresponding to the aqueous sample based on the extracted raw features;
generate classification information for bacteria species of the aqueous sample based on results of the machine learning operations and one or more of the excitation-emission matrix and the scattering spectrum of the aqueous sample; and
generate an output indicating the classification information for the bacteria species in the aqueous sample.
17 . The compute engine of claim 16 wherein the processing logic is further configured to:
measure an absorbance (A) corresponding to the selected wavelengths applied to the aqueous sample;
extract one or more raw features from the absorbance measured for the aqueous sample;
perform machine learning operations corresponding to the aqueous sample based on the extracted raw features;
generate classification information for bacteria species of the aqueous sample based on results of the machine learning operations and one or more of the excitation-emission matrix and the scattering spectrum of the aqueous sample; and
generate an output indicating the classification information for the bacteria species in the aqueous sample.
18 . The compute engine of claim 16 wherein the processing logic is further configured to:
extract at least one of an excitation (X) spectrum or an emissions (E) spectrum from the EEM;
extract one or more raw features from the excitation spectrum and/or the emissions spectrum measured for the aqueous sample;
perform machine learning operations corresponding to the aqueous sample based on the extracted raw features;
generate classification information for bacteria species of the aqueous sample based on results of the machine learning operations and one or more of the excitation-emission matrix and the scattering spectrum of the aqueous sample; and
generate an output indicating the classification information for the bacteria species in the aqueous sample.
19 . The compute engine of claim 16 wherein the selected subset of available wavelengths comprise 260 nm, 275 nm, 280 nm, 300 nm, 370 nm, 400 nm, 500 nm, and 600 nm.
20 . The compute engine of claim 16 wherein direct scattering measurements are supplemented with estimates of a spectra derivative by sampling incident and scattered light at λ exc −Δλ, λ exc and λ exc +Δλ, where λ exc is the nominal excitation wavelength and Δλ is the spectrometer resolution.