Hybrid and modular electronic nose

Abstract

The present invention provides a system (100) including a sensor unit (105) comprising a plurality of chemical sensors (113, 114, 115, 116), such as VOC sensors, a sensor unit circuitry (106, 107) using a pre¬ trained machine learning model to generate preprocessed chemical signals based on the chemical signals generated by said chemical sensors, and a first electrical connector (108) to releasably connect to a base unit connector (102) of a base unit (101). The system (100) further includes said base unit (101), which comprises said base unit connector (102), base unit circuitry (103) to identify a chemical based on the preprocessed chemical signals, and an output (104) to output the chemical identity.

Description

HYBRID AND MODULAR ELECTRONIC NOSECROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of US Provisional Application No. 63 / 622,960 filed on January 19, 2024, the contents of which are hereby incorporated in its entirety.BACKGROUND

[0002] There are correlations between the volatile organic compounds (VOCs) present in blood, the headspace of blood, breath, urine, sweat, or other bodily fluids, and various diseases, such as lung cancer, diabetes, kidney disease, and bacterial infections. Also, specific bacterial or viral infections can be detected by analyzing the odor profiles associated with these pathogens. Breath analysis has also shown potential in diagnosing or monitoring various diseases. By analyzing VOCs present in exhaled breath, certain diseases can be detected or their progression can be monitored. Also, this information can be used for personalized medicine, dietary monitoring, and disease management. Specifically, some alkanes can indicate the presence of lung tumors. Liver cancer changes the ammonia level in breath. An elevated level of nitric oxide (NO) and certain VOCs in a person's breath can be an indication of asthma. Scientific studies have also shown promise in detecting and monitoring diabetes by measuring the concentration of acetone and ethanol in a breath sample. Furthermore, the potential of detecting COVID by analyzing breath samples has been studied.

[0003] Although there is no doubt about the benefits of using an artificial nose for medical applications, designing an effective device is complicated. The olfactory sensory system in humans and some animals has a sophisticated design for distinguishing several thousands of different odors. The detection mechanism is based on the electrochemical interaction of odorant receptors (ORs) with the odor molecules. There are thousands of different ORs each with different sensitivities to different odors. However, the selectivity of ORs in responding to different gases / VOCs is limited. Hence, the signals from ORs are processed by the brain to combine the information from all ORs.BRIEF DESCRIPTION OF THE DRAWINGS

[0004] FIGS. 1A-C illustrate an example hybrid modular e-nose system.

[0005] FIGS. 2A-2F illustrate an example architecture and components of an example hybrid modular e-nose system.

[0006] FIGS. 3A-3B illustrate example test results of an example e-nose.

[0007] FIG. 4 illustrates an example method of applying a trained model to chemical data.

[0008] FIG. 5 illustrates an example method of training a model to analyze chemical data.

[0009] FIG. 6 illustrates an example computer system including an e-nose.

[0010] FIG. 7 illustrates an example design for a computer system including an e-nose.DETAILED DESCRIPTION

[0011] Endogenous VOCs are products of metabolic activity. In cancer, some metabolites increase due to tumor necrosis, resulting in cancer-specific VOC patterns that can be detected from bodily fluids. Blood plasma detection mitigates variations that can arise from the subject's activities or diet (which is a common issue when VOCs are measured in breath samples) and reduces interference in comparison to whole blood. For example, in patients with ovarian cancer (OC), analyses of specific VOCs in blood have shown promising outcomes in separating OC patients from healthy women. For various reasons, such as low concentrations of biomarkers (on a scale of sub-ng / mL), more sensitive sensors with higher selectivity towards the biomarkers and different algorithms are needed to improve the accuracy of the diagnostic process. As an example, the identified biomarkers in blood for diagnosing colorectal cancer (CRC) is challenging due to the relatively low sensitivity and selectivity of existing sensors in the market.

[0012] Aspects of the described technology may provide systems including chemical sensors, such as VOC sensors, and advanced artificial intelligence / machine learning (AI / ML) models, to provide a hybrid e-nose for CRC screening with a sensitivity>90% and specificity >90%. Although the CRC biomarkers are identified, it should be noted that cancer screening is very different than, for example, blood sugar monitoring. For sugar monitoring, there are sensors that are only sensitive to insulin, and it is known that 0.5-2.0 ng / mL of insulin in the blood is normal. The two major challenges for CRC screening are: a) There is no (convenient) method that can measure only the biomarkers and b) The specific concentration of each biomarker representing an unhealthy person is not known.

[0013] In general, both challenges can be addressed using e-nose technology. The described technology may use an array of different VOC sensors with an integrated circuit reading the sensors' signals and a code / algorithm to analyze the information to identify a pattern. Using AI / ML the e-nose can be trained to identify the patterns associated withvarious stages of cancer without a need for measuring the concentration of each VOC in the sample.

[0014] Mimicking the olfactory system, existing e-nose devices are essentially made of an array of different gas and VOC sensors with an integrated circuit reading the sensors' signals and a code / algorithm to analyze the information. The accuracy of detecting various odors depends on the number of sensors (and how the sensors are operated, which features are extracted for the training of the model), the sensors' selectivity and sensitivity, and the applied algorithm to interpret data. In contrast to several thousands of olfactory receptors (Ors) in the olfactory system, a typical commercial e-Nose may have 4-32 different sensor devices. Generally, current products with 32 sensors are expensive instruments with complex hardware and software (suitable for research) that are not practical for small hospita Is / cl i n ics or home point-of-care (POC) testing. The modular design of the hybrid e-Nose allows us to integrate medical instrument using different sensing technologies while offering an affordable solution by using fewer sensors each being more selective and sensitive.

[0015] Nanostructures of metal oxides (MOXs) such as ZnO, WO3, and SnCh have been used widely in the form of resistors, diodes, and transistors for making sensors used in e-nose devices. The dangling bonds at the surface of the materials allow the oxygen atoms to be reactive to the adsorbed chemicals. However, their relatively poor selectivity is a serious shortcoming not being able to distinguish for example various alcohols. More importantly, as our experimental result shows, due to the required high temperature operation (200-400C), MOX sensors are equipped with an internal heater that requires 20-30 min of warm-up time. This implies that a medical instrument which employs MOX sensors has to be operated in a ~30 min long warm-up cycle, prior to using it for an accurate reading.

[0016] FIG. 1 illustrates an example modular electronics chemical sensing system 100 for chemical analysis of gases or liquids ("e-nose"). E-nose 100 may be used for the detection and concentration measurements of different gases and chemicals in a mixed gas / liquid environment. For example, e-nose 100 may comprise a diagnostic tool capable of analyzing gases, volatile organic compounds (VOCs), etc.. In some case, e-nose 100 may comprise a medical diagnosis tool to analyze chemicals from breath, blood, urine, sweat, body odor, headspace of blood, wounds, etc. As another example, e-nose 100 may comprise an platform for hybrid sensors can also be used in the agriculture / food industries for product qualitycontrol, military / environmental engineering for hazardous chemicals detection, air quality monitoring, etc.

[0017] System 100 may include a sensor unit 105 that can be connected to and disconnected from a base unit 101. Sensor unit 105 may include various co-housed components, such as sensor array 109, circuitry 106, 107, and a connector 108. Sensor array 109 may comprise a plurality of sensors 113, 114, 115, 116 on an exposed surface of sensor unit 105 to measure chemical signals of gases / liquid disposed in a receptacle 110. For example, the receptacle 110 may comprise a breathalyzer comprising a mouthpiece 111 and exit port 112 with sensor array 109 disposed in a chamber (e.g., breath circulation chamber). In further examples, receptacle 110 may comprise an atmospheric gas capturing receptacle, a liquid receiving receptacle, an absorbant pad (e.g., to be pressed to a wound), etcSensor array 109 may include any number of chemical sensors 113-116. For example, sensor array 109 may comprise a grid of sensors 113-116 arranged in quadrants (as illustrated). Chemical sensors 113-116 may include any type of chemical sensor, such as, for example, chemiresistors, chemical sensitive field-effect transistors (chemFETs), chemcapacitors, electrochemical cells, etc. Chemical sensors 113-116 may include any suitable material, such as, for example, chemically sensitive resistive materials (e.g. carbon nanotubes) , metal-oxide semiconductors (MOXs) (e.g., ZnO, WO3, and SnCh), 2D materials (e.g., M0S2, functionalized MXene, graphene oxide (rGO)) metalorganic compounds (e.g., Metallophthalocyanines (MPcs) and metalloporphyrins (MPPs) (e.g., CuPc, CoPc, ZnPc, FePc, CuPP, CoPP, ZnPP, FePP), metal-organic frameworks (MOFs) (e.g., tailored structures such as zeolitic imidazolate framework-8 (ZIF-8)), covalent organic frameworks (COFs) (e.g., shaped COFs such as starshaped COFs, including perylene-based COFs such as Per-lP, Per-N, Per-Py COFs, etc)], nanodiamonds, organic semiconductors (e.g., polyaniline, polypyrrole, etc.), etc. Chemical sensors 113-116 may include any other materials / structures to support chemical sensing, such as conducting polymers or metal nanoparticles. Of course, these are simply examples, the described technology may accommodate any chemical sensing modality. In some examples, sensor array 109 may include other sensors 117, 118, such as a temperature sensor 117, a humidity sensor 118, etc. Data obtained from these other sensors 117, 118, may be used as data input for an AI / ML model to analyze chemical composition / concentrations, as calibration data, or for any other suitable use.

[0018] In some examples, system 100 may include multiple sensor units 105, 119, etc. For example, multiple sensor units 105, 119 may provide a modular platform for customizing the number of sensors to assemble a sensor set with hybrid sensors. For instance, the combination of the sensors may be changed to customize the hybrid e-Nose for sensing body odor, urine / blood, non-medical applications like air quality monitoring, etc. In some examples, base unit 101 may have a corresponding compatible set of sensor units 105, 119, etc. For instance, a set of seven sensor units 105, 119, each supporting a quad-sensor array, provides a group of 28 chemical sensors for chemical analysis. As an illustrative example, a set of sensor units may a set of seven sensor units having: 1) a first chemiresistor sensor array including CuPc, CoPc, ZnPc, FePc sensors; 2) a second chemiresistor sensor array including CuPP, CoPP, ZnPP, FePP sensors; 3) an electrochemical sensor array including ZIF sensors biased at four different voltages (e.g., -0.8V, -0.4V, 0.4V, 0.8V); 4) a FET sensor array including N-doped graphene biased at different voltages (e.g., 0.5V, 1.0V, 1.5V, 2.0V); 5) a first MOX sensor array (e.g., commercially available sensors TGS2600, TGS2602, TGS2610, TGS2611); 6) a second MOX sensor array (e.g., commercially available sensors ENS161, BME680, MICS5135, MICS5521); 7) a third MOX sensor array (e.g., commercially available sensors MQ-2, MQ-3, MQ-135, MQ-138). Accordingly, in various implementations, any battery of chemical tests may be supported. In some examples, each sensor unit has a corresponding sensor type (e.g., as illustrated in the previous seven module example) such has a particular sensor class or sensor material type.

[0019] In some examples, sensor groups may be provided to according to particular applications. For example, a sensor set for analyzing blood may be capable of distinguishing between VOC profiles pertaining to different conditions, such as type, subtype, and stage of cancers. As another example, as a breathalyzer, sensors units may be included that can detect important VOCs in human breath, including acetone, ethanol, acetic acid, isopropanol, NH3, NO, and CO. Further examples may include sensors sensitive to other compounds such as ethylbutanoate, acetaldehyde, propylacetate, etc. may also be found in human breath. Example sensor system 100 may sense concentrations as well as compositions. For instance, in a breathalyzer implementation 100, sensors with sensitivity and the minimum level of detection (LOD) at least 0.1 - 0.2 ppm may be included based on an approximate level of VOCs in human breath between 0.2 and 4.0 ppm.

[0020] Sensor units 105, 119 may include various circuitry 106 / 120, 107 / 121. For example, Circuitry 106 / 120 may include circuitry to support the operation and measurement for the sensors (e.g., analog front end circuitry, driver circuitry, etc.) to generate sensed chemical signals. Circuitry 107 / 121 may include machine learning circuitry to generate preprocessed chemical signals based on the sensed chemical signals. For example, circuitry 107 / 121 may include circuitry to generate calibrated chemical signals or circuitry to generate input features (e.g., latent vectors). For instance, circuitry 107 / 121 may comprise neuromorphic circuitry (e.g., an application specific integrated circuit (ASIC), field-programmable gate array (FPGA), resistive artificial neuron circuitry, photonic neural circuitry, etc.), an AI / ML processor, a microcontroller, data storage (e.g., memory, flash storage, etc.) storing a model, etc. For ease of explanation, circuitry 107 / 121 may be referred to as a microcontroller 107 / 121 (including a non-transitory computer readable medium to store instructions) with the understanding that the described technology is not so limited.

[0021] In some examples, circuitry 107 / 121 may preprocess the sensed chemical signals using a pre-trained machine learning model to generate calibrated chemical signals based on the sensed chemical signals. Due to the modular structure with each module (e.g., 4 sensors) equipped with a microcontroller 107 / 121, system 100 may include separate codes on each module 105, 119 for calibration. Calibration may be performed based on various data, such as temperature or humidity data received from sensors 117 / 118, age of the sensors (e.g., to compensate for sensor drift / loss of sensitivity), past usage of sensors (e.g., total time or times operated), air / gas pressure, flow-rates (e.g., of gas through a receptacle 110), or other data. For example, the circuitry 107 may implemented a calibration learning model via transfer learning machine learning approaches commonly (e.g., as used in image / video and language modeling) to reduce the learning load, improve the rate of training, and achieve a suitable end result (e.g., asymptotic training convergence). For example, the circuitry 107 / 121 may implement deep neural networks trained via reduced data sets to a provide a calibration process with no need to train on fresh samples. For example, in a medical deployment, In particular, calibration may be performed without using (fresh) patient samples. In some examples, other sensor units 105, 119 in a sensor unit group may have other calibration circuitry such as analog calibration circuitry 106, may lack calibration circuity, or may perform calibration via a microcontroller 107 / 121 executed other calibration techniques. In other examples, base unit 101 may perform sensor calibration for at least a portion of sensor units.

[0022] In some examples, circuitry 107 / 121 may preprocess chemical signals (e.g., sensed chemical signals, preprocessed chemical signals, etc.) to provide an input for base unit 101. For example, circuitry 107 / 121 may apply a dimensional reduction technique to the chemical signals. For example, circuitry 107 / 121 may apply principal component analysis (PCA) to chemical signal data. PCA is a dimensionality reduction technique that can be employed as a pre-processing step to simplify the data from various gas and VOC sensors. Since the hybrid sensing platform includes multiple sensors with different technologies, the data obtained from each sensor may have different dimensions and characteristics. PCA can help in reducing the feature space while preserving the most important patterns in the data. In this context, PCA can be applied to reduce the dimensionality of the raw sensor data, enabling easier handling of the data by subsequent ML models. By reducing the feature space, PCA can eliminate noise and redundancies, which may improve the overall performance of the subsequent ML algorithms.

[0023] As another example, circuitry 107 / 121 may apply linear discriminant analysis (LDA) to chemical signals (e.g., sensed signals, calibrated signals, PCA pre-processed signals, etc.). LDA is a supervised dimensionality reduction technique that aims to find a linear combination of features that best separates the classes (different gases / VOCs in this case). Given that the sensor array supports identifying the composition of target gases / VOCs, LDA can be applied to enhance the separation between different classes and improve the accuracy of the classification. After using PCA for dimensionality reduction, LDA can be applied to further optimize the feature space by maximizing inter-class separability while minimizing intra-class variance. This step can help enhance the discrimination between different gases / VOCs, thus improving the performance of the ML model. For instance, circuitry 107 / 121 may comprise a variational autoencoder (VAE), multilayer perceptron, or other circuitry to generate an embedding into a latent space.

[0024] As a further example, circuitry 107 / 121 may generate a data representation of signals to be input into a subsequent model. For example, circuitry 107 / 121 may encode signals into model feature data, latent vectors, embedding vectors, etc. As an example, circuitry 107 / 108 may generate between 70-150 (e.g., 80-100) different feature values representative of chemical signals. As another example, circuitry 107 / 121 may generate a spatial representation of received data. For example, circuitry 107 / 121 may process chemical signals to generate a 2D (e.g., from a representative sensed output, from a statisticalcombination of time-varying signals, etc..) or 3D spatial / spatiotemporal dataset (e.g., a series of formatted spatial data representing sensed signals over time). For instance, circuitry 107 / 121 may comprise transform circuitry (e.g., fast-Fourier transform (FFT), etc.) or formatting circuitry to generate a spatial representation (a "chemical image") with respect to received signals.

[0025] In various examples, circuitry 107 / 121 may generate separate preprocessed chemical signals corresponding to each of the sensors on array 109, 123. Circuitry 107 / 121 may also generate preprocessed chemical signals based on a combination of multiple sensor data. For instance, circuitry 107 / 121 may generate chemical images for each of the sensors 113-116 separately or may combine signals from sensors 113-116 to generate chemical images.

[0026] Sensor units 105, 119 may include an electrical connector 108 / 122 to connect to the base unit 101 (e.g., at a base unit connector 102). For example, the connector 108 / 122 may include a USB connector, an M.2 connector, a special-purpose connector, general input / output (GPIO) connector, etc. to provide a data bus or other channel to base unit circuitry 103. In some examples, electrical connector 102 / 122 may comprise an electromechanical connector to provide both a mechanical connection and an electrical connection. In other examples, sensor units 105, 119 may mechanically connect to base unit 101 in any removable manner. For example, base unit 101 may comprise a physical socket for the sensor units 105, 119, sensor units 105, 119 may slidably attach with base unit 101 (e.g., via keyways and detents, etc.), etc. In some examples, sensor unit 105 / 119 may further be powered by base unit 101. Sensor unit 105 / 119 may join a voltage domain of the base unit via the connector (e.g., similar to RAM modules), base unit 101 may output a power signal (e.g., via a USB connection), etc. In still other examples, sensor unit 105 / 119 may be separately powered (e.g., via battery, solar power, etc.).

[0027] Base unit 101 may comprise base unit circuitry 103 to identify a chemical based on the preprocessed chemical signals. For example, circuitry 103 may comprise circuitry to apply an AI / ML model to preprocessed chemical signals received from a sensor unit 105 / 119. As an example, circuitry 103 may comprise a microcontroller or other embedded processor coupled to a non-transitory computer readable media. As another example, circuitry 103 may comprise neuromorphic circuitry, a system-on-a-chip, etc. As an example, circuitry 103 may comprise a microcontroller and a storage storing a convolutional neural network (CNN). CNNsare powerful deep learning models commonly used for image recognition tasks. To leverage the potential of CNNs, the responses from the multiple sensors can be arranged in a structured format, similar to image data, and used as input to the CNN model. As discussed above, sensor unit 105 / 119 may provide spatially formatted data that can be treated as a "chemical image," where the responses of various gas and VOC sensors represent the spatial information of the gases / VOCs present in the sample (e.g., breath). In other examples, circuitry 103 may include circuitry to format the preprocessed data, including as instructions stored on a medium and executed by a processor. The CNN can be trained to learn complex patterns and relationships among different gases / VOCs, which can be crucial for accurate identification and concentration estimation. The CNN model can be implemented with any neural network design, including multiple convolutional layers, followed by pooling and fully connected layers for feature extraction and classification. The model's output layer can provide the predicted composition of the chemicals and their corresponding concentrations.

[0028] Combining PCA, LDA, and CNN in the proposed AI / ML model can help address the challenges posed by the hybrid sensing platform, where multiple types of sensors with diverse characteristics are used. PCA and LDA can aid in data preprocessing and feature optimization, while CNN can capture complex patterns in the structured sensor data, leading to more accurate and precise predictions of the gases / VOCs' composition and concentrations.

[0029] Of course, any suitable AI / ML model may be used, including neural networks, support vector machines, random forests, reservoir computers, etc. Different applications may of course use different model types, for instance given the complex nature of VOC patterns in cancer detection, deep learning models have shown promise, as they can capture intricate relationships within the data.

[0030] In some examples, the AI / ML model embodied in circuitry 103 may operate on data combined from multiple sensor units 105 / 119. For example, a user may couple a first sensor unit 105 to base unit 101 and take a first measurement, then exchange the first sensor unit 105 for a second sensor unit 119 and take a second measurement, which may continue to obtain a particular battery of sensor measurements. For instance, a 28 sensor example (based on 4 sensors per unit with 7 units) was discussed above. In that example, preprocessed signals corresponding to all 28 sensors might be applied to an AI / ML model to identify chemicals and concentrations of those chemicals.

[0031] In some examples, base unit 101 may be configurable with a particular AI / ML model. For instance, an AI / ML model may be installed on device storage in circuitry 103 to be executed by a microcontroller. As another example, an AI / ML model may be instantiated via programming the connectivity of an FPGA or other programmable neuromorphic circuitry. For example, a base unit 101 may be updated with an updated AI / ML model or parameters of an existing model may be updated.

[0032] In some examples, sensor units 105 / 119 may be useable with multiple different base units 101 for different types of chemical detection. For example, a sensor unit 105 for detecting chemical implicated in different cancers may be useable on different base units 101 dedicated to particular cancer detection models. In some cases, sensor unit circuitry 106 / 107 may include circuity to identify a particular base unit 101 and to reconfigure itself. For example, connector 102 might include an identity chip or dedicated identity pin that may be detected by a sensor unit 105 / 119 to implement a particular configuration. For instance, sensor units 105 / 119 might operate at different sampling rates, bias voltages, heater temperatures, etc.

[0033] Base unit 101 may further comprise output circuitry 104 to output signals, such as results of the AI / ML analysis, sensor data (e.g., a copy of the preprocessed data received from a sensor unit), telemetry, etc. Output circuitry 104 may include a removable storage device, a port for a wired connection, a wireless transceiver, etc. For example, output 104 may comprise a wireless transceiver for a communication protocol such as WiFi, Bluetooth, 5G / 6G cellular communications, body area network communications, etc. Base unit 101 may provide output data to a local device or a server (e.g., a cloud location), such as responsive to a received request, automatically in response to completing an analysis, on a schedule, etc.

[0034] FIGS. 2A-2F illustrate various aspects of an architecture 200 for an e-nose. For example, various aspects of architecture 200 may be implementations of a system 100.

[0035] In this example an e-nose device 205 may include one or more modular sensor arrays 201. For example, device 205 may include a chemiresistors array 204, a FET array 208, and an electrochemical sensor array 211.

[0036] A chemiresistor array 204 may comprise a plurality of chemiresistors 203, which may include various sensing materials 202. For example, chemiresistors 203 may include metalloporphyrins (MPPs), such as CuPP, MnPP, MgPP, ZnPP, LiPP, etc.; metallophthalocyanines (MPcs), such as CuPc, MnPc, MgPc, ZnPc, LiPc; functionalized ordoped materials, such as functionalized / doped graphene, M0S2, Mxene, carbon nanotubes, etc.;; metal oxides (MOXs), such as ZnO, WO3, and SnCh, etc.; and other suitable chemiresistor materials [[are there other classes of chemiresistor) ; etc. As discussed above, in some examples, an array 204 may comprise sensors with materials within the same class (e.g., one array 204 might comprise MPPs and another might comprise MPcs, etc.), while in others, an array 204 may comprise sensors with materials from different classes.

[0037] A FET sensor array 208 may comprise a chemically sensitive conductive material 206 coupled to terminals (e.g., a gate terminal G, a drain terminal D, and a source terminal S), where binding of the molecules to the senso surface alters the conductivity of the material 206. For example, FET sensor materials may include 2D materials such as graphene or M0S2, etc; ID materials such as silicon nanowires, carbon nanotubes, etc.; organic semiconductors; metal oxides; lll-V semiconductors; transition metal dichalcogenides (e.g. WS2, WSe2, VTe2, etc.); etc.

[0038] An electrochemical sensor array 211 may comprise a plurality of electrochemical sensors 210, which may include various sensing materials 209. For example, electrochemical sensors 210 may include metal-organic frameworks (MOFs), such as MOFs with a zeolite carrier (e.g., zeolitic imidazolate), etc. Of course, these are simply examples, and the technology may utilize any sensor modality. Indeed as new chemical sensing modalities are developed, sensor units may be developed to be incorporated into existing systems.

[0039] As discussed above, the e-nose 205 may apply an AI / ML model 212 to determine a chemical makeup 213 of an environment. For instance, the illustrated graph shows detection and concentrations of NH3, ethanol, isopropanol, CO, acetic acid, NO, and acetone, such as may be detected from the breath of a subject. The analysis results 213 may be transmitted 214 to an external computer system 215 (e.g., a medical office server, laptop, etc.) for further processing. The outputs of the AI / ML models from the hybrid sensors may serve as a transfer learning for various applications, such as health monitoring, diagnostics, environmental sensing, air quality, explosives detection, etc. As an example, a diagnostic application could e.g. be a test designed to determine the presence of a certain condition. For example, output of e-nose 205 may be used as an input to a binary classifier AI / ML model (e.g., with other subject data such as health history, health status, etc.). In some examples, the e-nose 205 may implement a binary classifier model to identify the presence of a condition. For example, circuitry 103 may include a stored binary classification model that may be applied to datareceived from sensor units. As another example, health monitoring applications may utilize a more complex model, such as to determine concentrations to make informed decisions, e.g. about medicine dosage. Applications, depending on specifics, may also require training the model with weights favoring one or several metrics while penalizing others, depending on which health technology metric is prioritized, e.g. tuning the position on a ROC (receiver operating characteristic) curve to favor a high true positive.

[0040] Accordingly, an AI / ML model framework can be developed for automatic feature selection, redundancy reduction, and optimization of hyperparameters while re-training the model for specific diagnostic and health monitoring applications. Feature redundancy reduction serves two purposes: 1. it reduces the risk of 'overfitting', which can occur when the number of observations is lower or near the number of features. 2. It results in a more robust algorithm / classification, as the most likely component to drift / fail are the individual sensors; fewer features - fewer sensors - a more robust system. K-fold cross-validation (with the K number selected based on the data set) and (when the data set permits) leave one out cross-validation will be conducted to evaluate the performance of the e-Nose towards the specific applications. This ensures that the models are tested to make predictions on unknown data.

[0041] FIG. 2B illustrates an example chemical sensor quad array 208. In particular, quad array 208 comprises a FET array comprising FET sensor 207 as described above. In this example, each sensor 207 comprises a corresponding analog circuit 216 providing an interface to receive signals and control functions (e.g., analog circuits 216 may comprise implementations of circuitry 106 as described above). Circuits 216 may provide various functionality, such as noise mitigation circuits, pre-amplification circuits, bias control circuits, etc. Quad array 208 further comprises a microcontroller 220 comprising a processor and associated storage (e.g., an embedded SoC, etc.). For example, microcontroller 220 may comprise an implementation of circuitry 107 as described above. Sensor array modules, such as sensor array 208, may comprise a circuit board or other circuit substrate having sensors 207 disposed on a first (exposed) side and circuitry 216, 217 exposed on an opposite (unexposed) side. Alternatively, or additionally, sensor array 208 may comprise a semiconductor wafer having sensors 207 on top (exposed) side and having circuitry 216, 220 on the back side. For example, sensor array 208 may be manufactured using SoC technology, including wafer bonding, etc. FIG. 2C illustrates an example of a hybrid sensor platform. Here,multiple sensor units (represented by corresponding sensor arrays 204, 208, 211) may be used in a common base unit 214 to obtain any number of desired sensor readings. While illustrated as quad arrays, sensor arrays 204, 208, 211 may have any number of sensors. For example, with the number of sensors may be based on balancing the traffic of signal communication with the main processor and flexibility in the modular design.

[0042] FIG. 2D illustrates an example chemiresistor sensor 204 including interface circuitry. In this example, chemiresistor 203 comprises a resistor of a voltage divider including second resistor 227 and port 221. The output of the voltage divider is an input to an opamp 222. A DC offset is provided to the port of opamp 222 (e.g., provided by a digital signal by a microcontroller coupled to a digital to analog converter (DAC) 223). The output of opamp 222 is provided to a digitizer 224 (e.g., an analog to digital converter (ADC)). In operation, the resistance of chemiresistor 203 changes based on the presence / absence of the chemical(s) to which the material 203 is sensitive. The change in resistance changes the voltage output 221 by the voltage divider 227, 203. In this example, a second ADC 225 provides a measurement prior to chemical exposure that is compared to the output of ADC 224 to provide a chemical signal. While illustrated as a second ADC 225, in some examples, the pre-measurement reading may be performed by the first ADC 224 prior to chemical exposure.

[0043] FIG. 2E illustrates an example FET sensor 208 including interface circuitry. In this example, FET 207 is operated in active mode (e.g., between cutoff and saturation) so that drain current is responsive to the presence of the target chemical(s). A first bias voltage may be provided for a drain-source bias, such as from a DAC 235 or other voltage course. Similarly, a second bias voltage may be provided for a gate-source bias, such as from a second DAC 230. The output current of FET 207 is input to an opamp 226 with a negative feedback resistor 228 (e.g., for gain control). The current is compared to ground and the output is provided to ADC 229 which outputs a digitized chemical signal.

[0044] FIG. 2F illustrates an example electrochemical sensor 211 with chronoamperometry circuitry. In this example, a DAC 235 provides a constant cell bias voltage to current electrode of an electrochemical cell 210. A reference voltage and working electrode current is provided to the inputs of an opamp 234, respectively to output a feedback signal via ADC 236. A cell current is measured via an opamp 232 with negative feedback via resistor 233 and a positive input connected to ground. The output of opamp 232 is digitized via ADC 237. The output signals of ADCS 236, 237 may be compared to determine the chemical signal.

[0045] The inventors have studied four specific MPcs (CuPc, CoPc, ZnPc, and FePc). The samples were made by drop casting a solution containing the chemicals on commercially available interdigitated gold electrodes to fabricate chemiresistors. FIG. 3A illustrates an example response of a CuPc chemiresistor to various VOCs. As an example, the results from CuPc and a commercial MOX sensor (ENS161) are shown in FIG. 3B when they were exposed to NH3. The results clearly show two orders of magnitude changes in the resistivity of CuPc while the MOX sensor (presumably very sensitive to NH3) presented only ~10% change in the resistance. Also, testing CuPc with a few VOCs (FIG. 3A), a very selective response was achieved when being exposed to formic acid. Testing the sensors with mixed gases show a pattern consistent with the response from the tests with single VOCs.

[0046] Referring now to FIG. 4, a flowchart is illustrated as setting forth the steps of an example method for using a suitably trained neural network or other machine learning model, such as, for example, the neural networks described above. In some examples, the neural network or other machine learning model takes chemical data (e.g., sensed chemical signals) as input data and generates preprocessed chemical data as output data. In other examples, the neural network or other machine learning model takes preprocessed chemical data as input data and generates a chemical analysis as output data. In still further examples, the neural network or other machine learning model takes chemical analysis data as input data and generates a diagnosis or health information as output data.

[0047] The method includes accessing chemical data (e.g., raw data, preprocessed data, analysis data, etc.) with a computer system, as indicated at step 402. Accessing the chemical data may include retrieving such data from a memory or other suitable data storage device or medium. Additionally or alternatively, accessing the chemical data may include acquiring such data with a chemical sensor and transferring or otherwise communicating the data to the computer system, which may be a part of an e-nose.

[0048] A trained neural network (or other suitable machine learning model) is then accessed with the computer system, as indicated at step 404. In general, the neural network is trained, or has been trained, on training data. This evaluation is achieved, in part, by the neural network (or other machine learning model) being trained via a chemical dataset. For example, the training dataset may comprise chemical data captured from blood, breath, etc., which may be annotated with the ground truth measures. For example, training the models may use blood samples from biobanks.

[0049] The trained neural network can include a neural network with any suitable neural network architecture for generating chemical data outputs (e.g., preprocessed chemical data, chemical analysis data, health / diagnosis data, etc.). As one non-limiting example, the trained neural network or machine learning model may include a convolutional neural network, deep neural network, reservoir computer, decision tree forest, etc. The trained neural network may in some instances have multiple inputs (e.g., corresponding to chemical data from multiple sensors).

[0050] Accessing the trained neural network may include accessing network parameters (e.g., weights, biases, or both) that have been optimized or otherwise estimated by training the neural network on training data. In some instances, retrieving the neural network can also include retrieving, constructing, or otherwise accessing the particular neural network architecture to be implemented. For instance, data pertaining to the layers in the neural network architecture (e.g., number of layers, type of layers, ordering of layers, connections between layers, hyperparameters for layers) may be retrieved, selected, constructed, or otherwise accessed. In some examples, accessing the trained neural network may comprise operating a neural network instantiated on an e-nose (e.g., a sensor unit or base unit). The chemical signal data are then input to the trained neural network, generating output as described above.

[0051] The output chemical data generated from applying the input chemical data to the trained neural network(s) can then be provided to a user, stored for later use or further processing, or both, as indicated at step 408. For example, the chemical data may be stored on an e-nose base unit, may be transmitted to an attached device, uploaded to a server, such as a clinical computer system (e.g., a server at a clinic), displayed on the e-nose (e.g., via an indicator light to indicate a health status output), etc.

[0052] Referring now to FIG. 5, a flowchart is illustrated as setting forth the steps of an example method for training one or more neural networks (or other suitable machine learning models) on training data, such that the one or more neural networks are trained to receive chemical data as input data in order to generate output chemical data. For example, in some implementations, neural network training may be performed on a computer system (e.g., system 600) prior to being instantiated on an e-nose device (e.g., one or more devices of system 100).

[0053] In general, the neural network(s) can implement any number of different neural network architectures. For instance, the neural network(s) could implement a convolutional neural network, a residual neural network, or the like. Alternatively, the neural network(s) could be replaced with other suitable machine learning or artificial intelligence algorithms, such as those based on supervised learning, unsupervised learning, deep learning, ensemble learning, dimensionality reduction, and so on.

[0054] The method includes accessing training data with a computer system, as indicated at step 502. In general, the training data can include chemical data with ground truth annotations generated from input chemical data. Additionally or alternatively, the accessed training data can include chemical data received from an example database. Accessing the training data may include retrieving such data from a memory or other suitable data storage device or medium. Alternatively, accessing the training data may include acquiring such data with a chemical sensing device and transferring or otherwise communicating the data to the computer system.

[0055] The method can include assembling training data from chemical data using a computer system. This step may include assembling the chemical data into an appropriate data structure on which the neural network or other machine learning model can be trained. Assembling the training data may include annotating chemical data. For instance, assembling the training data may include measuring or obtaining chemical signals, annotating the chemical data with known ground truth values, preprocessing chemical data, and the like.

[0056] One or more neural networks (or other suitable machine learning models) are trained on the training data, as indicated at step 504. In general, the neural network can be trained by optimizing network parameters (e.g., weights, biases, or both) based on minimizing a loss function. As one non-limiting example, the loss function may be a mean squared error loss function. Other tests may be applied during training. For example, to align with standard health technology metrics in oncology, ML model performance metrics, including receiver operating characteristic (ROC) curves, validation and test confusion matrices, sensitivity, specificity, and F-score values may be evaluated. Additionally, feature importance analysis may be conducted using chi-square, F-test, and the ReliefF algorithm.

[0057] Training a neural network may include initializing the neural network, such as by computing, estimating, or otherwise selecting initial network parameters (e.g., weights, biases, or both). During training, an artificial neural network receives the inputs for a trainingexample and generates an output using the bias for each node, and the connections between each node and the corresponding weights. For instance, training data can be input to the initialized neural network, generating output as chemical data. The artificial neural network then compares the generated output with a ground truth value of the training example in order to evaluate the quality of the chemical output data. For instance, the chemical data can be passed to a loss function to compute an error. The current neural network can then be updated based on the calculated error (e.g., using backpropagation methods based on the calculated error). For instance, the current neural network can be updated by updating the network parameters (e.g., weights, biases, or both) in order to minimize the loss according to the loss function. The training continues until a training condition is met. The training condition may correspond to, for example, a predetermined number of training examples being used, a minimum accuracy threshold being reached during training and validation, a predetermined number of validation iterations being completed, and the like. When the training condition has been met (e.g., by determining whether an error threshold or other stopping criterion has been satisfied), the current neural network and its associated network parameters represent the trained neural network. Different types of training processes can be used to adjust the bias values and the weights of the node connections based on the training examples. The training processes may include, for example, gradient descent, Newton's method, conjugate gradient, quasi-Newton, Levenberg-Marquardt, among others.

[0058] The artificial neural network can be constructed or otherwise trained based on training data using one or more different learning techniques, such as supervised learning, unsupervised learning, reinforcement learning, ensemble learning, active learning, transfer learning, or other suitable learning techniques for neural networks. As an example, supervised learning involves presenting a computer system with example inputs and their actual outputs (e.g., categorizations). In these instances, the artificial neural network is configured to learn a general rule or model that maps the inputs to the outputs based on the provided example input-output pairs.

[0059] The one or more trained neural networks are then stored for later use, as indicated at step 506. Storing the neural network(s) may include storing network parameters (e.g., weights, biases, or both), which have been computed or otherwise estimated by training the neural network(s) on the training data. For example, storing the neural network may include instantiating the neural network on a sensor unit, or a base unit of an e-nose, such byprogramming a neuromorphic computer or storing neural network parameters in a device storage system, controller, etc. Storing the trained neural network(s) may also include storing the particular neural network architecture to be implemented. For instance, data pertaining to the layers in the neural network architecture (e.g., number of layers, type of layers, ordering of layers, connections between layers, hyperparameters for layers) may be stored.

[0060] FIG. 6 shows an example of a system 600 for evaluating chemicals in accordance with some embodiments described in the present disclosure. As shown in FIG. 6, a computing device 650 can receive one or more types of data (e.g., chemical signal data, including raw chemical signals and / or preprocessed chemical data) from data source 602. For example, computing device 650 may comprise an e-nose and data source 602 may comprise a chemical sensor array. In some embodiments, computing device 650 can execute at least a portion of a chemical identification process 601 to generate health data (e.g., respiratory health data) from data received from the data source 602.

[0061] Additionally or alternatively, in some embodiments, the computing device 650 can communicate information about data received from the data source 602 to a server 652 over a communication network 654, which can execute at least a portion of the chemical identification process 601. In such embodiments, the server 652 can return information to the computing device 650 (and / or any other suitable computing device) indicative of an output of the chemical identification process 601. In some embodiments, data source 602 can be any suitable source of data (e.g., recorded chemical data, measured chemical data, preprocessed chemical data, etc.), such as a chemical sensor, another computing device (e.g., a sensor unit including preprocessing circuitry), and so on.

[0062] In some embodiments, communication network 654 can be any suitable communication network or combination of communication networks. For example, communication network 654 can include a Wi-Fi network (which can include one or more wireless routers, one or more switches, etc.), a peer-to-peer network (e.g., a Bluetooth network), a cellular network (e.g., a 3G network, a 4G network, etc., complying with any suitable standard, such as CDMA, GSM, LTE, LTE Advanced, WiMAX, etc.), other types of wireless network, a wired network, and so on. In some embodiments, communication network 654 can be a local area network, a wide area network, a public network (e.g., the Internet), a private or semi-private network (e.g., a corporate or university intranet), any other suitable type of network, or any suitable combination of networks. Communicationslinks shown in FIG. 6 can each be any suitable communications link or combination of communications links, such as wired links, fiber optic links, Wi-Fi links, Bluetooth links, cellular links, and so on.

[0063] Referring now to FIG. 7, an example of hardware 700 that can be used to implement data source 602 (e.g., a sensor unit), computing device 650 (e.g., a base unit), and server 652 in accordance with some embodiments of the systems and methods described in the present disclosure is shown.

[0064] As shown in FIG. 7, in some embodiments, computing device 650 can include a processor 702, a display 704, one or more inputs 706, one or more communication systems 708, and / or memory 710. In some embodiments, processor 702 can be any suitable hardware processor or combination of processors, such as a central processing unit (CPU), a graphics processing unit (GPU), and so on. In some embodiments, display 704 can include any suitable display devices, such as a liquid crystal display (LCD) screen, a light-emitting diode (LED) display, an organic LED (OLED) display, an electrophoretic display (e.g., an "e-ink" display), a touchscreen, and so on. In some embodiments, inputs 706 can include any suitable input devices and / or sensors that can be used to receive user input, such as a keyboard, a touchscreen, a microphone, and so on.

[0065] In some embodiments, communications systems 708 can include any suitable hardware, firmware, and / or software for communicating information over a data bus, communication network 654 and / or any other suitable communication networks. For example, communications systems 708 can include one or more transceivers, one or more communication chips and / or chip sets, and so on. In a more particular example, communications systems 708 can include hardware, firmware, and / or software that can be used to establish a Wi-Fi connection, a Bluetooth connection, a cellular connection, an Ethernet connection, and so on.

[0066] In some embodiments, memory 710 can include any suitable storage device or devices that can be used to store instructions, values, data, or the like, that can be used, for example, by processor 702 to present content using display 704, to communicate with server 652 via communications system(s) 708, and so on. Memory 710 can include any suitable volatile memory, non-volatile memory, storage, or any suitable combination thereof. For example, memory 710 can include random-access memory (RAM), read-only memory (ROM), electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM), other formsof volatile memory, other forms of non-volatile memory, one or more forms of semi-volatile memory, one or more flash drives, one or more hard disks, one or more solid state drives, one or more optical drives, and so on. In some embodiments, memory 710 can have encoded thereon, or otherwise stored therein, a computer program for controlling operation of computing device 650. In such embodiments, processor 702 can execute at least a portion of the computer program to present content (e.g., images, user interfaces, graphics, tables), receive content from server 652, transmit information to server 652, and so on. For example, the processor 702 and the memory 710 can be configured to perform the methods described herein.

[0067] In some embodiments, server 652 can include a processor 712, a display 714, one or more inputs 716, one or more communications systems 718, and / or memory 720. In some embodiments, processor 712 can be any suitable hardware processor or combination of processors, such as a CPU, a GPU, and so on. In some embodiments, display 714 can include any suitable display devices, such as an LCD screen, LED display, OLED display, electrophoretic display, a computer monitor, a touchscreen, a television, and so on. In some embodiments, inputs 716 can include any suitable input devices and / or sensors that can be used to receive user input, such as a keyboard, a mouse, a touchscreen, a microphone, and so on.

[0068] In some embodiments, communications systems 718 can include any suitable hardware, firmware, and / or software for communicating information over communication network 654 and / or any other suitable communication networks. For example, communications systems 718 can include one or more transceivers, one or more communication chips and / or chip sets, and so on. In a more particular example, communications systems 718 can include hardware, firmware, and / or software that can be used to establish a Wi-Fi connection, a Bluetooth connection, a cellular connection, an Ethernet connection, and so on.

[0069] In some embodiments, memory 720 can include any suitable storage device or devices that can be used to store instructions, values, data, or the like, that can be used, for example, by processor 712 to present content using display 714, to communicate with one or more computing devices 650, and so on. Memory 720 can include any suitable volatile memory, non-volatile memory, storage, or any suitable combination thereof. For example, memory 720 can include RAM, ROM, EPROM, EEPROM, other types of volatile memory, other types of non-volatile memory, one or more types of semi-volatile memory, one or more flashdrives, one or more hard disks, one or more solid state drives, one or more optical drives, and so on. In some embodiments, memory 720 can have encoded thereon a server program for controlling operation of server 652. In such embodiments, processor 712 can execute at least a portion of the server program to transmit information and / or content (e.g., data, images, a user interface) to one or more computing devices 650, receive information and / or content from one or more computing devices 650, receive instructions from one or more devices (e.g., a personal computer, a laptop computer, a tablet computer, a smartphone), and so on.

[0070] In some embodiments, the server 652 is configured to perform the methods described in the present disclosure. For example, the processor 712 and memory 720 can be configured to perform the methods described herein (e.g., the method of FIG. 4, the method of FIG. 5).

[0071] In some embodiments, data source 402 can include a processor 722, one or more data acquisition systems 724 (e.g., sensors), one or more communications systems 726 (e.g., a connector to a base unit 650), and / or memory 728. In some embodiments, processor 722 can be any suitable hardware processor or combination of processors, microcontroller, SoC„ and so on. In some embodiments, the one or more data acquisition systems 724 comprise chemical sensors to acquire data, such as from a liquid or gas. In some embodiments, one or more portions of the data acquisition system(s) 724 can be removable and / or replaceable.

[0072] In some embodiments, communications systems 726 can include any suitable hardware, firmware, and / or software for communicating information to computing device 650 (and, in some embodiments, over communication network 654 and / or any other suitable communication networks). For example, communications systems 726 can include one or more transceivers, one or more communication chips and / or chip sets, and so on. In a more particular example, communications systems 726 can include hardware, firmware, and / or software that can be used to establish a wired connection using any suitable port and / or communication standard (e.g., DVI video, USB, RS-232, etc.), Wi-Fi connection, a Bluetooth connection, a cellular connection, an Ethernet connection, and so on.

[0073] In some embodiments, memory 728 can include any suitable storage device or devices that can be used to store instructions, values, data, or the like, that can be used, for example, by processor 722 to control the one or more data acquisition systems 724, and / or receive data from the one or more data acquisition systems 724; to generate images from data; present content (e.g., data, images, a user interface) using a display; communicate withone or more computing devices 650; and so on. Memory 728 can include any suitable volatile memory, non-volatile memory, storage, or any suitable combination thereof. For example, memory 728 can include RAM, ROM, EPROM, EEPROM, other types of volatile memory, other types of non-volatile memory, one or more types of semi-volatile memory, one or more flash drives, one or more hard disks, one or more solid state drives, one or more optical drives, and so on. In some embodiments, memory 728 can have encoded thereon, or otherwise stored therein, a program for controlling operation of data source 402. In such embodiments, processor 722 can execute at least a portion of the program to generate images, transmit information and / or content (e.g., data, images, a user interface) to one or more computing devices 650, receive information and / or content from one or more computing devices 650, receive instructions from one or more devices (e.g., a personal computer, a laptop computer, a tablet computer, a smartphone, etc.), and so on.

[0074] In some embodiments, any suitable computer-readable media can be used for storing instructions for performing the functions and / or processes described herein. For example, in some embodiments, computer-readable media can be transitory or non- transitory. For example, non-transitory computer-readable media can include media such as magnetic media (e.g., hard disks, floppy disks), optical media (e.g., compact discs, digital video discs, Blu-ray discs), semiconductor media (e.g., RAM, flash memory, EPROM, EEPROM), any suitable media that is not fleeting or devoid of any semblance of permanence during transmission, and / or any suitable tangible media. As another example, transitory computer- readable media can include signals on networks, in wires, conductors, optical fibers, circuits, or any suitable media that is fleeting and devoid of any semblance of permanence during transmission, and / or any suitable intangible media.

[0075] As used herein in the context of computer implementation, unless otherwise specified or limited, the terms "component," "system," "module," "framework," and the like are intended to encompass part or all of computer-related systems that include hardware, software, a combination of hardware and software, or software in execution. For example, a component may be, but is not limited to being, a processor device, a process being executed (or executable) by a processor device, an object, an executable, a thread of execution, a computer program, or a computer. By way of illustration, both an application running on a computer and the computer can be a component. One or more components (or system, module, and so on) may reside within a process or thread of execution, may be localized onone computer, may be distributed between two or more computers or other processor devices, or may be included within another component (or system, module, and so on).

[0076] In some implementations, devices or systems disclosed herein can be utilized or installed using methods embodying aspects of the disclosure. Correspondingly, description herein of particular features, capabilities, or intended purposes of a device or system is generally intended to inherently include disclosure of a method of using such features for the intended purposes, a method of implementing such capabilities, and a method of installing disclosed (or otherwise known) components to support these purposes or capabilities. Similarly, unless otherwise indicated or limited, discussion herein of any method of manufacturing or using a particular device or system, including installing the device or system, is intended to inherently include disclosure, as embodiments of the disclosure, of the utilized features and implemented capabilities of such device or system.

Claims

CLAIMSWhat is claimed is:

1. A system, comprising: a sensor unit, comprising: a plurality of chemical sensors to generate sensed chemical signals; sensor unit circuitry to process the sensed chemical signals using a pre-trained machine learning model to generate preprocessed chemical signals based on the sensed chemical signals; and a first electrical connector to releasably connect to a base unit connector; the base unit, comprising: the base unit connector to releasably connect to the first electrical connector; base unit circuitry to identify a chemical based on the preprocessed chemical signals; and an output to output the chemical identity.

2. The system of claim 1, further comprising: a second sensor unit, comprising: a second plurality of chemical sensors to generate second sensed chemical signals; second sensor unit circuitry to process the second sensed chemical signals using a second pre-trained machine learning model to generate second preprocessed chemical signals based on the second sensed chemical signals; and a second electrical connector to releasably connect to the base unit connector; wherein the base unit circuitry is to identify the first chemical or a second chemical based on the second preprocessed chemical signals.

3. The system of claim 2, wherein: the base unit circuitry is to generate spatially structured representations of data received from the first sensor unit and the second sensor unit; and the base unit circuitry comprises circuitry to apply a pretrained convolutional neural network (CNN) model to the spatially structured representations.

4. The system of claim 3, wherein the first sensor unit circuitry is to apply a first dimensional reduction process to generate the data received from the first sensor unit and the second sensor unit circuitry is to apply a second dimensional reduction process to generate the data received from the second sensor unit.

5. The system of claim 2, wherein the first plurality of chemical sensors are of a first class of chemical sensors and the second plurality of chemical sensors are of a second class of chemical sensors.

6. The system of claim 5, wherein at least one of the first class of chemical sensors or the second class of chemical sensors comprises: metal-oxides, metal organic frameworks (MOFs), covalent organic frameworks (COFs), metallophthalocyanines (MPcs), metalloporphyrins (MPPs), nanodiamonds, MXene, Graphene, M0S2, or composites thereof.

7. The system of claim 5, wherein first class of chemical sensors comprises one of chemiresistors, field effect transistors (FETs), or electrochemical cells; and wherein the second class of chemical sensors comprises another one of chemiresistors, field effect transistors (FETs), or electrochemical cells.

8. The system of claim 1, wherein the sensor unit circuitry is to process the sensed chemical signals to generate feature data and the base unit circuitry is to process the feature data using a second pre-trained neural network to identify the chemical.

9. The system of claim 1, wherein: the sensor unit circuitry comprises a sensor unit processor and a sensor unit non- transitory computer readable medium storing the pretrained machine learning model and instructions executable by the sensor unit processor to apply the pre-trained machine learning model to generate the preprocessed chemical signals; and the base unit circuitry comprises a base unit processor and a base unit non-transitory computer readable medium storing instructions executable by the base unit processor to identify the chemical.

10. The system of claim 9, wherein the instructions executable by the base unit processor to identify the chemical are modifiable to identify the chemical based on second preprocessed chemical signals received from a second sensor unit releasably connected to the base unit connector.

11. The system of claim 1, wherein the sensor unit comprises a circuit board comprising the plurality of chemical sensors disposed of a first side of the circuit board, and at least a portion of the sensor unit circuitry disposed on a second side of the circuity board.

12. The system of claim 1, wherein the plurality of chemical sensors comprises four chemical sensors arranged in respective quadrants of a sensing surface.

13. The system of claim 1, further comprising: a temperature sensor; and a humidity sensor; wherein the pre-trained machine learning model is to generate the preprocessed chemical signals based on temperature data and humidity data.

14. The system of claim 13, wherein the sensor unit comprises the temperature sensor and the humidity sensor.

15. The system of claim 1, wherein the output comprises a transceiver to transmit the output.

16. The system of claim 15, further comprising a computer comprising: a second transceiver to receive the output; a processor; and a non-transitory computer readable medium storing instructions executable by the processor to: apply a second pretrained machine learning model to the output to generate health information for a subject.

17. The system of claim 1, wherein at least one of the plurality of chemical sensors comprises a chemiresistor, a field effect transistor (FETs), or an electrochemical cell.

18. The system of claim 1, wherein at least one of the plurality of chemical sensors comprises a metal-oxides, a metal organic frameworks (MOF), a covalent organic framework (COFs), a metallophthalocyanine (MPc), a metalloporphyrin (MPP), a nanodiamond, an MXene, graphene, M0S2, or a composite thereof.

19. A sensor unit device, comprising: a plurality of chemical sensors to generate sensed chemical signals; sensor unit circuitry to process the sensed chemical signals using a pre-trained machine learning model to generate preprocessed chemical signals based on the sensed chemical signals; and a first electrical connector to releasably connect to a base unit connector.

20. The sensor unit device of claim 18, wherein the sensor unit circuitry is to apply a dimensional reduction process to generate data to output to the base unit.

21. The sensor unit device of claim 18, wherein: the sensor unit circuitry comprises a sensor unit processor and a sensor unit non- transitory computer readable medium storing the pretrained machine learning model and instructions executable by the sensor unit processor to apply the pre-trained machine learning model to generate the preprocessed chemical signals.

22. The sensor unit device of claim 18, further comprising a circuit board comprising the plurality of chemical sensors disposed of a first side of the circuit board, and at least a portion of the sensor unit circuitry disposed on a second side of the circuity board.

23. The sensor unit device of claim 18, further comprising: a temperature sensor; and a humidity sensor; wherein the pre-trained machine learning model is to generate the preprocessed chemical signals based on temperature data and humidity data.

Images