A cold plasma sensor for minimally destructive detection of biological materials

Abstract

A biologic tissue identification and differentiation system includes a cold- atmospheric plasma device having a plasma discharge end, the plasma end to direct plasma at a biologic sample, an optical sensor to capture optical characteristics from the plasma and the biologic sample, an electrical sensor to capture electrical characteristics from one or more electrodes and the biologic sample, and a machine learning system to receive at least one of the optical characteristics and the electrical characteristics to output a classification of a tissue type of the biological sample. A method of identifying biologic tissue includes applying a cold-atmospheric plasma to a biologic sample, capturing one or more optical characteristics from the plasma and the biologic sample, capturing one or more electrical characteristics from one or more electrodes and the biologic sample, and using a machine learning system to receive at least one of the optical characteristics and electrical characteristics to output a tissue type classification of the biological sample.

Description

A COLD PLASMA SENSOR FOR MINIMALLY DESTRUCTIVE DETECTION OF BIOLOGICAL MATERIALSTECHNICAL FIELD

[0001] This disclosure relates to cold atmospheric plasmas, more particularly to cold atmospheric plasma devices configured to detect and differentiate biological materials.BACKGROUND

[0002] A major challenge in cancer diagnosis involves the early identification and differentiation between healthy and malignant tissue. The primary tool for cancerous cell determination is the histological evaluation (i.e., examination of the cell structures under microscope) of an extracted piece of tissue. This is an invasive procedure that involves the physical removal of potentially afflicted tissue (via surgery) and is not done in real-time. Particularly in the skin cancer realm, a non-invasive and real-time procedure to identify and discriminate cancerous tissues would significantly benefit patients. Creating one would allow for earlier diagnosis and can help avoid potentially unnecessary surgical procedures that can have complications and that are time consuming. A few alternatives have been proposed recently but are not widely used due to various limitations still involving the need for expert analysis.

[0003] One such example involves the use of high-frequency ultrasounds to obtain a high- resolution image of the skin structure via sound wave reflection measurements. Another very recent example involves the use of optical coherence tomography (OCT) to perform a “virtual” biopsy. In this “virtual” biopsy, the procedure involves a non-invasive threedimensional scan of the tissue. The scan is used to generate a stained tissue image, similar to that of a true stained sample, which may be used instead of a true stained sample to providediagnostic insights. To replicate and potentially replace expert analysis for real-time and early diagnosis, more recently, there have been advances in artificial intelligence (Al) for medical image analysis. The accuracy of Al-based automatic detection and diagnostic systems has been shown to be comparable to that of experienced physicians in radiology with the turnaround time significantly improved, which can pose a potential avenue to speed up cancer diagnoses. In fact, medical image analysis using Al for skin cancer detection has been proposed in several studies, but this method still lacks reliability since it relies purely on visual appearance of the afflicted area. Thus, this method still relies on subjective characteristics of the macro-scale morphology of the tissue. Cancerous tissues, meanwhile, are known to have physiochemical properties that are significantly different from non- cancerous / healthy tissue.

[0004] Since the physiochemical properties of different tissue types play an important role in the existing highest standard (i.e., histological analysis) toolkit, it must also be part of any new and innovative approach, and a few such properties have been explored. For example, one method proposed this based on the measurement of electrical conductivity and capacitance of melanoma cells. Their technique required the use of disposable gold needles in an 8x8 matrix designed to penetrate the stratum comeum in the tested skin area and was calibrated based on surrounding healthy tissue. This technique showed high sensitivity (92% compared to 75% for physicians), but low specificity (67% compared to 87% for physicians) for the diagnosis of melanoma. Another method relied on analyzing the chemical composition of cancerous tissue. This method used optical emission spectroscopy of an atmospheric thermal plasma (with temperatures that exceed 1000°C) generated using an electrosurgical tool used during a tumor resection. The tissue in contact with the thermal plasma is vaporized allowing for the excitation of the molecular makeup of the tissue, which emits characteristic optical emission spectra that can discriminate cancerous versus non-cancerous tissue. This method reported an accuracy of 95% in differentiating between cancerous and healthy liver tissues. However, it is destructive and not applicable in skin cancer detection and early diagnosis. Finally, tumor cells have been reported to present lower thermal conductivity compared to healthy cells, yet no known cancer detection system based on thermal conductivity has been proposed. To summarize, key limitations to each of the above methods include the focus on a single property, the lack of specificity, and / or the destructive nature of the examination.BRIEF DESCRIPTION OF THE DRAWINGS

[0005] FIG. 1 shows a schematic of an embodiment of a plasma gun setup, a data collection setup, and examples of plasma gun interaction with biologic tissue.

[0006] FIG. 2 shows a graphical representation of information flow for an embodiment of a tissue diagnostic process.

[0007] FIG. 3 shows exemplary optical emission spectra of helium plasma impinging upon skin, muscle, bone, and fat tissue of biological model.

[0008] FIGs. 4A-4B show exemplary charge-voltage figures of helium plasma impinging upon skin, muscle, bone, and fat tissue of biological model.

[0009] FIGs. 5A-5B show confusion matrices having three subplots for embodiments of different machine learning systems trained on different data sets.

[0010] FIGs. 6A-D show selected biologically relevant peaks from optical emission spectra for cold-atmosphere plasma-treated biological models.DETAILED DESCRIPTION OF THE EMBODIMENTS

[0011] The embodiments here present a new approach to real-time tissue differentiation using cold atmospheric plasmas (CAPs) to provide insight into various physiochemical properties of biological tissue. CAPs are a form of partially ionized gaseous matter that can be generated at near-room temperature and atmospheric pressure, which allows for a “gentle”interaction with biological materials, such as tissues. This “gentle” interaction means that CAPs can be used in a non-invasive, minimally destructive manner. Thus, CAPs have sparked the development of an entirely new field that lies at the intersection of (nonequilibrium) plasma and medicine. Plasma medicine has now grown into a field that ranges in proposed treatments from disinfection to wound healing to cancer therapies. The key to CAPs’ success in medicine has been the plasma-surface interactions with biological systems, which can be evaluated by monitoring the physiochemical properties of those interactions. However, plasma medicine hinges on expertise in plasma physics and chemistry, as well as in the biochemistry of the plasma interface interactions. Al and machine learning (ML) can play a crucial role in elucidating the underlying physics of plasma-interface interactions in realtime. Recently, there have been some reports on learning-based control for CAPs, on ML for predicting biological outcomes of CAP treatments, and on predicting physiochemical properties to differentiate (non-biological) materials. However, no known works use CAPs to differentiate biological targets.

[0012] This work aims to use a combination of CAP effects and ML to detect differences in, differentiate between, and diagnose / identify biological tissues. The embodiments investigate how CAP interactions with biological tissues can be combined with ML to identify biological tissues in a real-time, non-invasive manner. To do so, the inventors developed and used an all-in-one CAP generation and CAP effect measurement device. The CAP device is based on a prior configuration commonly proposed for plasma medicine combined with an automated data acquisition system using real-time capable measurement devices commonly used in plasma (and plasma-surface) characterization. This setup was used to collect chemical and electrical data on ex vivo chicken leg models at various points consisting of different tissue types (i.e., skin, muscle, bone, and fat). The chemical data primarily consisted of opticalemission spectra, and the electrical data consisted of electrical waveforms associated with the plasma-target interactions.

[0013] Once data were collected with the setup, physics- and biologically informed data analysis and processing were used to identify and select features of the data to be used in training ML models. This data processing includes, but is not limited to, peak selection to identify differences in chemical properties, and physical transformation of the data to understand underlying physics. After this physics-informed data processing, the inventors evaluated a variety of supervised learning classification techniques to achieve high test accuracy across all models.

[0014] FIG. 1 shows a schematic of the CAP device and data acquisition setup in FIG. 1. The CAP device configuration, developed at the GREMI laboratory and known as the “Plasma Gun,” is a coaxial dielectric barrier discharge jet using helium 10 as the working gas. High voltage ps pulses (+7 kV peak amplitude with 600 Hz pulse repetition frequency using a custom power supply 14) are applied to an enclosed brass electrode 12, and an outer electrode 16 surrounding the quartz capillary tube 20 serves as a ground electrode. Once the high voltage pulse is applied, helium flowing (maintained at 0.5 SLM via a Bronkhorst EL-FLOW Prestige FG-201 CV) through the quartz capillary tube is ionized. The ionization front propagates outside the tube partially ionizing the surrounding ambient air to form a “flame” 21. The plasma gun maintains a distance of 5 mm shown by arrow 22 between the tip of tube and the biological interface. One should note that the below discussion involves a cold atmospheric plasma jet, but any cold atmospheric plasma reactor that generates plasma from an electrode, in a microwave cavity or by means of induction coupling is within the scope of the embodiments.

[0015] A fiber optic cable 20, 45° from the tube axes and pointed at the plasma-tissue incidence point, is connected to a spectrometer or other optical sensor 26, and used to collectoptical emission spectra. A compensation circuit 33 is used to mimic the electrical interactions with a non-human material to that of a human interface. In use, the sensor would reside on the connection that connects the patient to ground. The discussion here focuses on a spectrometer, but the system may use other types of optical sensors that can capture optical emission spectra or other optical characteristics of which optical emission spectra is one. In addition, other types of plasma reactor configurations would be possible. This particular cold plasma jet provides one example. Also, the system discussed here may be combined with other types of diagnostic systems such as hyperspectral cameras, ultraviolet light-visible (UV-Vis) detectors including UV-Vis spectrometers, cameras, etc.

[0016] One should note that the position of the optical sensor may change as the user decides. For example, rather than being positioned at the plasma / tissue interaction point, the optical sensor could be directed at the plasma itself. Similarly, the electrodes may reside in several places in addition to, or instead of, the locations shown in the embodiment of FIG. 1. For example, the electrode may reside as the plasma excitation / generation point, and after the plasma / tissue interaction point.

[0017] In the embodiment shown in FIG. 1, electrical characteristics of this system were taken at the locations marked by pentagons H, G, and T, numbered respectively as 30, 32 and 34, with voltage probes connected to an electrical sensor 36, which may comprise an oscilloscope. The electrical characteristics are taken from one or more of the electrodes and the biologic sample. The discussion here focuses on an oscilloscope and waveforms, but the system may use other types of electrical sensors and electrical characteristics, of which waveforms are one. This discussion will refer to the electrical sensor 36 as an oscilloscope and the optical sensor 26 as a spectrometer for ease of discussion and without limitation. In the experiments, the inventors used a raw chicken leg model to test various biological tissues, and the chicken leg in the schematic is for illustrative purposes only. The spectrometer 26 andoscilloscope are connected to a computer 40 to automatically obtain and record data at 0.5 s sampling intervals.

[0018] As will be discussed in more detail later, the computing device may contain or have access to a machine learning system. In some embodiments, processor 40 may execute code to use machine learning to analyze the results from the sensors. Alternatively, the processor may communicate with another computing device that provides access to a machine learning system.

[0019] One should note that the below discussion involves particular devices, machine learning models, and implementations of the machine learning models. These are used for the ease of discussion and are not intended to limit any of the components of the system to any particular device or model.

[0020] In an experiment, two resolutions of spectra were acquired with the Plasma Gun. The lower resolution optical emission spectra (OES) were taken using an Ocean Optics Maya 2000 Pro connected to a UV-VIS fiber optic cable. The integration time was 200 ms with a spectral resolution of about 0.4 nm. The high-resolution OES were acquired with the same optical fiber setup connected to a 0.320 m focal length spectrometer (IsoPlane SCT 320) coupled with an ICCD camera (PiMax4 by Princeton Instruments). Here, the ICCD camera was synchronized with the high-voltage pulse of the CAP device and the total integration time was 10 ps per 5000 on-chip accumulations. Low-resolution OES were used in the data- driven classification, and the high-resolution OES were used to validate physical findings between the spectral differences observed in the low-resolution data.

[0021] Electrical characteristics were taken using voltage probes connected to a PicoScope 2406B. The voltage probes were located at positions specified using pentagons in FIG. 1. The applied high-voltage measurement was taken at location A, 30, using a high-voltage probe.The voltage probes used at the ground electrode, location B, 32 and after the compensation circuit, location C, 34, may differ from the high-voltage probe used at A.

[0022] Classification is a category of supervised ML wherein a model predicts the “label” of given input data. In this work, this reflects the correct prediction of the type of biological model tissue based on chemical and electrical input data. FIG. 2 illustrates the information flow of the proposed classification technique. The system captures and saves raw chemical data, in the form of OES, and electrical data from the Plasma Gun. The system processes the data in a second step, where it selects peaks of the OES that indicated plasma-biological significance and discards the remainder of the spectra.

[0023] Electric waveforms were processed into Lissajous curves. The process labels each of these types of processed data with the tissue type of the sample, skin, muscle, bone, and fat, and uses it to train and evaluate various ML models. Once a model is trained, the first two steps can be re-used during real-time inference to generate a prediction of the tissue type, as illustrated in the final step of FIG. 2. In addition to the type of tissue identifiable by this system, skin, muscle, fat, and bone, the system may also identify cancerous and non- cancerous tissue. For purposes of this discussion, the Lissajous figures are considered to be a type of electrical waveform to be sent to the machine learning system. One should note that the discussion involves particular methods to process raw data signals from particular devices and is used for ease of discussion. They are not intended to limit the scope of the embodiments to any one or few selected data processing strategies.

[0024] The embodiments herein underwent investigation on the value of each type of data, only chemical, only electrical, or a combination of chemical and electrical, for tissue identification by training ML models for each type of model input. Furthermore, the inventors compared the performance of several types of machine learning models for multiclass classification, including ^ nearest neighbor, decision trees, random forests, andneural networks. Each of k nearest neighbors, decision trees, and random forests was created and trained using the scikit-leam. Fully connected neural networks were created and trained using Tensorflow.

[0025] The machine learning models above comprise supervised learning models. In some embodiments, the data from the optical and electrical sensors may first go through an unsupervised learning preprocessing such as dimensionality reduction or clustering as examples.

[0026] Plasma diagnostics commonly use OES; however, the embodiments here involve the first uses of this information for the characterization of the plasma-(bio)interface interactions. FIG. 3 illustrates an exemplary OES of the helium plasma impinging upon different tissues of a chicken leg model, FIG. 3 shows the broader wavelength range in the main frame, and the inset shows a larger view of a shorter range. Using the automatic data acquisition setup, the inventors obtained OES of the CAP interactions with skin, muscle, fat, and bone tissue of ex vivo chicken leg models over a treatment time of 50 seconds at 0.5-second sampling intervals. The helium CAP shows characteristic peaks for helium excitation (marked He) at 587.6 nm, 667.8 nm, 706.5 nm, and 728.1 nm. Further, oxygen and nitrogen species typical of CAP operating in ambient air include O (777 nm), N2 (315.93, 337.13, 357.69, and 380.49 nm), NH (336.01 nm), NO (236 nm and 248 nm), HNO2 (354.25 nm), N2+(358.21, 391.44, and 427.81 nm). One can see noticeable differences among the relative intensity of these peaks for CAP interacting with different tissues, suggesting that plasma chemistry can be influenced through interactions with different biological interfaces.

[0027] Several spectral regions showed differences between the various substrates, but their identification requires further investigation as the literature involving OES with biomolecules is extremely limited and associated with other methods, such as laser-induced breakdown spectroscopy (LIBS), thermal plasma and light-induced fluorescence (LIF).

[0028] Lissajous curves provide visualizations of a system of parametric equations, typically of two waveforms. In the case of CAP systems, charge-voltage Lissajous figures can be generated from measurements of the applied voltage and the charge deposited to target and electrodes. Charge is measured via the voltage drop through a capacitor. In creating such figures for the CAP system, physics information is encoded into a visual representation of the data, in particular, the area enclosed by the Q-V figure is the amount of energy deposited onto the target electrode.

[0029] FIGs. 4A-4B illustrate exemplary charge-voltage (Q-V or Lissajous) figures to demonstrate the electrical properties of the plasma-tissue interactions. FIGs. 4A-4B illustrate the Q-V figures of measurements of the CAP system at two locations from FIG. 1 : one probe connected at 32, the outer grounded electrode of the CAP device shown in FIG. 4A; and one probe connected in series with the biological tissue after the compensation circuit, 34 shown in FIG. 4B. The inventors collected electrical waveforms over 50 seconds at 0.5 second sampling intervals and averaged the waveforms. The figures have a solid line showing the mean and three standard errors around the mean in light shading. Note that, in this example, the variation within each tissue is small such that there is no apparent shading in FIG. 4A, and that the variation in the fat tissue sample is small such that there is no apparent shading.

[0030] Qualitatively, it is possible to observe how the Q-V plot for the target shown on the right is clearly affected by the type of tissue. FIG. 4B illustrates different shapes of the electrical characteristics between different tissue types. This difference allows for differentiation or identification of biological tissues. Contrarily, the Q-V plot for the ground electrode in FIG. 4A remains essentially constant, having overlapped electrical characteristics and illustrates that the generated plasma has consistent behavior. This is because the internal configuration of the CAP device, where the plasma ignites, is fixed. One can assume that the plasma ignition inside the device is not significantly affected by the presence of the tissuecompared to the plasma’s propagation and interactions with the target tissues. The variation of the Q-V plot for the target can be exploited in differentiating or identifying biological tissues.

[0031] Following the data processing steps outlined in FIG. 2, the process trained several ML models to classify four different tissue types (skin, muscle, bone, fat) of ex vivo chicken leg models. Using the Plasma Gun setup, the inventors obtained 11,456 samples collected at 100 time intervals over seven chicken leg models at four different tissue types at multiple locations. Note that some data were manually excluded when the system did not ignite the plasma or acquire the data. Data was split with an 80 / 20 training / test split where 20% of the total samples were reserved as unseen data for evaluation. The remaining 80% was split into a 92 / 8 training / validation split, where 8% of the training data would be reserved for validation.

[0032] Each of the decision tree (DT), random forest (RF), and k nearest neighbor (kNN) models were trained using the fit function, and optimal hyperparameters of max tree depth and number of neighbors were selected via a 5-fold cross-validation score. The neural network (NN) was constructed as a fully connected network with three hidden layers, 100 nodes per layer, a batch normalization layer after the input, and a dropout layer (with 40% dropout) prior to the output layer. The NN was trained using a batch size of 128 over 30 epochs, and the best model was selected according to the best validation loss. Table 1 reports the test accuracy of these ML models trained on the chemical (OES) and electrical (Q-V image of the target) data. Accuracies are reported as the mean and confidence bounds (standard deviation) over 9 random initializations of each model. All models show high discriminative capabilities (> 90% test accuracy on average).Table 1

[0033] FIGs. 5A-B show examples of confusion matrices of classifiers trained on different sets of chicken model data. A confusion matrix is a common visual representation to illustrate the accuracy of a classification model. In a confusion matrix, the model predictions are plotted along one axis, while the true labels are plotted along the orthogonal axis. This results in a c x c grid, where c is the number of classes and the diagonal indicates when the prediction is equivalent to the truth. Hence, higher values along the diagonal of the confusion matrix indicate a higher accuracy. Further, the confusion matrix provides a representation of what “confuses” the model, i.e., what the model’s incorrect predictions should be in truth.

[0034] Specifically, FIG. 5 A illustrates the predictive capability of a DT, and FIG. 5B shows a NN using chemical-only data, electrical-only data, and a combination of both. In general, using chemical data results in higher tissue classification accuracy, as indicated by most sample predictions lying on the diagonal in both types of models when chemical data is included. Furthermore, a lower complexity model, such as a DT, can experience a performance boost, meaning more samples along the diagonal, when incorporating additional data compared to a higher complexity model, such as an NN. This performance boost is illustrated in FIG. 5 A when the DT increases in total number of samples along the diagonal, meaning accuracy: 1910 (83.4%) for the electrical data versus 2072 (90.4%) for the chemical data versus 2118 (92.4%) for the combined electrical and chemical data. Shown in FIG. 5B, the NN does not necessarily experience this performance boost, particularly when comparingthe use of chemical data versus combined data: 1919 (83.8%) for the electrical data versus2288 (99.9%) for the chemical data versus 2268 (99.0%) for the combined electrical and chemical data.

[0035] The embodiments demonstrated the potential for CAPs as a diagnostic tool for the identification and classification of biological tissues. CAPs exhibit distinct chemical and electrical characteristics when subjected to different biological materials. In particular, OES, which captures the gas-phase chemical reactivity of the plasma at the plasma-tissue incidence point, illustrates distinctions in the reach on / activati on potentials of both plasma and biological species. In particular, certain peaks of the OES can correspond to the excitation of compounds found in different amounts in each tissue type. The biological tissue classification may be applied to different types of tissues, such as the bone, muscle, skin, and fat discussed here, as well as skin and other types of cancers and infections as other examples. Further, no limitation exists, nor should any be implied, as to the nature of the biological sample. The biological sample may be human or may be animal if this system were employed in veterinarian environments.

[0036] CAPs exhibit distinct chemical and electrical characteristics when subjected to different biological materials. In particular, OES, which captures the gas-phase chemical reactivity of the plasma at the plasma-tissue incidence point, illustrates distinctions in several spectral regions shown in FIGs. 6A-6D. One way that the OES can capture distinctions between tissues may be due to the differences in water content among the investigated tissue types (i.e., skin 60-76%, muscle 73-78%, bone 44-55%, fat 5-20%). Thus, the quenching of helium metastable by water molecules and consequential decrease in nitrogen ion levels would also vary among the tissue types. The concentration of biomolecules such as collagen and elastin, commonly found in the investigated tissues, is also known to greatly vary among tested tissues. These or other biomolecules may be associated with the modulation of otherspectral regions, such as shown by the varying wavelengths in FIGs. 6A-6D. Even without a complete spectral peak identification, results support the evidence that observable differences in various spectral regions (especially molecular bands of nitrogen and nitrogen ions) captured the distinctions of different tissues such that the ML models were able to classify different tissue types with high accuracy. Given the potential of OES capabilities in this form of tissue differentiation, it would be prudent to investigate the sensitivity of the OES and devise data processing strategies such as normalization to further improve and standardize the method. Doing so may not only increase the classification capability, i.e., accuracy, but also reduce the necessary training resources, such as data, time, and / or model complexity.Table 2

[0037] The embodiments introduce a new method for real-time tissue identification using cold atmospheric plasmas and ML. In an experiment, inventors collected chemical, in the form of optical emission spectra, and electrical, in the form of circuit analysis, data from minimally destructive plasma interactions with ex vivo chicken leg models. Data were captured for distinct tissue types and were shown to be discriminative across tissue types, which can be attributed to distinct physiochemical properties that are exposed when different tissue types interact with the CAP. As such, various ML models were able to classify tissue types of ex vivo chicken leg models from features including the chemical and electrical data with at least 90% test accuracy on average and up to 99.5% test accuracy for the best model. The embodiments have identified and transformed the data into physically- and biologicallyrelevant features and elaborated on several components that may have led to such high predictive power.

[0038] Additionally, this written description makes reference to particular features. It is to be understood that the disclosure in this specification includes all possible combinations of those particular features. For example, where a particular feature is disclosed in the context of a particular aspect, that feature can also be used, to the extent possible, in the context of other aspects.

[0039] Also, when reference is made in this application to a method having two or more defined steps or operations, the defined steps or operations can be carried out in any order or simultaneously, unless the context excludes those possibilities.

[0040] All features disclosed in the specification, including the claims, abstract, and drawings, and all the steps in any method or process disclosed, may be combined in any combination, except combinations where at least some of such features and / or steps are mutually exclusive. Each feature disclosed in the specification, including the claims, abstract, and drawings, can be replaced by alternative features serving the same, equivalent, or similar purpose, unless expressly stated otherwise.

[0041] Although specific aspects of this disclosure have been illustrated and described for purposes of illustration, it will be understood that various modifications may be made without departing from the spirit and scope of the invention. Accordingly, the invention should not be limited except as by the appended claims.

Claims

WHAT IS CLAIMED IS:

1. A biologic tissue identification and differentiation system, comprising: a cold-atmospheric plasma device having a plasma discharge end, the plasma end configured to direct plasma at a biologic sample adjacent the discharge end; an optical sensor having a fiber optic cable positioned to capture optical characteristics from the plasma and the biologic sample; an electrical sensor positioned to capture electrical characteristics from one or more electrodes and the biological sample when the plasma is directed at a biological sample; and a machine learning system to receive at least one of the optical characteristics and the electrical characteristics to output a classification of type of tissue of the biological sample.

2. The biologic tissue identification and differentiation system as claimed in claim 1, further comprising one or more additional detectors to capture thermal, chemical, and electrical interactions of plasma with the incident biologic tissue.

3. The biologic tissue identification and differentiation system as claimed in claim 2, wherein the additional detector comprises one or more of a hyperspectral camera, an ultraviolet light visible spectrometer, and an ultraviolet light camera.

4. The biologic tissue identification and differentiation system as claimed in claim 1, wherein the machine learning system output comprises a classification of skin, muscle, fat, and bone.

5. The biologic tissue identification and differentiation system as claimed in claim 1 further comprising a computing device having one or more processors configured to execute code to convert the electrical characteristics in the form of electrical waveforms to Lissajous figures and to send the electrical waveforms to the machine learning system as Lissajous figures.

6. The biologic tissue identification and differentiation system as claimed in claim 1, further comprising a computing device having one or more processors configured to execute code to reduce a number of peaks of optical emission spectra as the optical characteristics by retaining only those peaks that indicate plasma-biological significance and to send the peaks that indicate plasma-biological significance to the machine learning system.

7. The biologic tissue identification and differentiation system as claimed in claim 1, wherein the machine learning system comprises a supervised learning model comprising one of a decision tree, a random forest, k nearest neighbor, and a neural network.

8. A method of identifying biologic tissue, comprising: applying a cold-atmospheric plasma to a biologic sample; capturing one or more optical characteristics from the plasma and the biologic sample; capturing one or more electrical characteristics when the plasma is directed at a biological sample from one or more electrodes and the biological sample; and using a machine learning system to receive at least one of the optical characteristics and the electrical characteristics to output a classification of a type of tissue of the biological sample.

9. The method as claimed in claim 8, wherein the machine learning system output comprises a classification of skin, muscle, fat, and bone.

10. The method as claimed in claim 8, wherein the one or more electrical characteristics comprise waveforms.

11. The method as claimed in claim 10, further comprising converting the electrical waveforms to Lissajous figures and to send the electrical waveforms to the machine learning system as Lissajous figures.

12. The method as claimed in claim 8, wherein the one or more optical characteristics comprise optical emission spectra.

13. The method as claimed in claim 12, further comprising reducing a number of peaks of the optical emission spectra by retaining only those peaks that indicate plasma-biological significance and to send the peaks that indicate plasma-biological significance to the machine learning.

14. The method as claimed in claim 8, wherein using the machine learning system comprises using a supervised learning model comprised of such as a decision tree, a random forest, k nearest neighbor, and a neural network.

15. The method as claimed in claim 8, wherein using a machine learning system comprises using unsupervised learning to preprocess data and then using the preprocessed data in a supervised learning process of tissue identification and differentiation.

Images