Systems, methods and devices for screening a subject for disease - Patents.com

Abstract

A system, method and device for screening a subject for disease is provided. The system comprises a portable medical screening device including a collection chamber for collecting a test sample of the subject, a set of chemically sensitive sensors, an analysis chamber configured to assign the set of chemically sensitive sensors, the set of chemically sensitive sensors adapted to detect VOCs in the test sample and generate an output signal indicative of the presence or absence of VOCs in the test sample as a result of the detection, and a first processing unit operably connected to the set of chemically sensitive sensors to receive the generated output signal. The system also comprises artificial intelligence-based classification software for determining a result related to the disease by processing and classifying the generated output signal. The classification software can be executed by the first processing unit, a software application installed on a computing device, or a remote processing unit.

Description

[Technical field]

[0001] The present invention relates generally to cancer screening solutions, and more particularly to non-invasive and non-irradiating disease screening systems, devices and methods. [Background technology]

[0002] As early as Roman times, medicine paid special attention to human physiological metabolic products; for example, uncontrolled diabetes was historically diagnosed by a sweet taste in the urine, and liver failure produced a fishy odor. However, human metabolic research did not meet the oncology field until April 1989, when Dr. Hywel Williams and Dr. Andres Pembroke of King's College Hospital, London, reported a case in The Lancet about a Collie-Doberman owner who attended their practice. She claimed that her dog had shown an increasing interest in licking a mole on her leg. The mole was then shown to be cancerous and was removed, saving the patient's life.

[0003] This turning point revealed that cancer produces metabolic changes in human physiology, thus altering the body's taste, texture and smell. Since then, several publications have been published that aim to find specific molecules or metabolites in biological samples that alert the presence of cancer in the body, i.e. cancer biomarkers. Examples of such studies include the search for lung cancer biomarkers in patients' breath, ovarian cancer biomarkers in urine and indeed breast cancer biomarkers in urine.

[0004] The prior art has one common feature: the publications are typically based on the principle that the presence of certain metabolites in a test sample is associated with the diagnosis of a disease, e.g., breast cancer. However, the described invention focuses not only on the presence of certain components in a sample, but also on the proportional relationship between them, as well as the comparison between the recent sample and previously analyzed and diagnosed samples.

[0005] Similarly, the American Cancer Society predicts that 30% of all cancers diagnosed in the United States in 2020 will be breast cancer. Breast cancer is considered to be the leading cause of death in women. However, research dedicated to it is not proportionate to its incidence. In fact, the NIH has recognized that women are underrepresented in medical research. This trend can be observed in the field of oncology, especially in light of current breast cancer screening. In fact, a study by the Centers for Disease Control and Prevention (CDC) in the United States stated that only 65% ​​of women attended the center in the past two years, and that one-third of breast cancers may have been detected too late, thus worsening the women's prognosis and chances of survival. One of the many factors that increases the chances of curing the disease is its early detection. It is very important to detect the presence of the disease at an early stage that helps in its management and cure.

[0006] Mammography is currently the gold standard for breast cancer screening, using low-energy X-rays to examine the human breast for diagnosis and screening. However, although the radiation dose is minimal, continuous exposure to mammograms can cause cancer. Furthermore, 46% of mammogram no-shows are due to pain. Furthermore, the Catalan Healthcare System reported that 93% of positive results are false positives. All these drawbacks, summarized in that mammography is painful, irradiating, and insensitive, drive the need for improved methods for breast cancer diagnosis.

[0007] Additionally, there are other breast cancer screening solutions available on the market, such as magnetic resonance imaging (MRI) and ultrasound. However, neither of these two image-based techniques can detect all types of breast cancer. In fact, image-based techniques often face difficulties in distinguishing breast tumors from healthy dense tissue in women with fibrocystic breasts (more than 40% of women worldwide actually have fibrocystic breasts). Finally, genetic testing can also diagnose the probability of having certain types of breast cancer, but this method is very expensive.

[0008] The present invention is intended to ameliorate the shortcomings of mammography screening methods and other cancer screening solutions available on the market.

[0009] Other methods and apparatus in this field are known from the following patents and patent applications: US Pat.

[0010] WO2019095790A1 discloses a method and apparatus for detecting and identifying various types of gas molecules or volatile organic compounds (VOCs) that are indicative of disease. A sensor array comprising a plurality of sensing and reference electrodes is used to measure an electrical signal indicative of the presence or concentration of VOCs from an exhaled breath sample. An algorithm processes the data collected from the array and generates a response pattern that is compared to a response pattern from a healthy subject.

[0011] US Patent No. 8,366,630 (B2) relates to a system for detecting VOCs from breath samples, comprising an apparatus comprising an array of chemically sensitive sensors of single-walled carbon nanotubes coated with non-polar small organic molecules, and a processing unit comprising a learning and pattern recognition analyzer, which receives the sensor output signals and compares them with stored data.

[0012] US Patent Application Publication No. 20120326092(A1) discloses a method for diagnosing colon cancer in a test subject by collecting a test breath sample from the test subject, determining a level of at least one VOC from a set of VOCs indicative of colon cancer in the test sample, and comparing the level of the at least one VOC from the test sample with the level of the at least one VOC in a control sample, wherein a significantly different level of the at least one VOC in the test sample compared to the level of the compound in the control sample is indicative of colon cancer.

[0013] US Patent Application Publication No. 2019271685(A1) relates to a system including a selected, well-defined sensor set including at least three sensors responsive to the presence of VOCs in the breath of a test subject, where the sensors include nanomaterials selected from metal nanoparticles coated with a first organic coating and single-walled carbon nanotubes (SWCNTs) coated with a second organic coating, and a processing unit including a pattern recognition analyzer, where the pattern recognition analyzer receives, for example, an output signal of the sensor set.

[0014] US Patent No. 10011481 (B2) provides an electronic olfactory device based on a specific chemically sensitive field effect transistor. In particular, the sensor of the electronic olfactory device is composed of non-oxidized functionalized silicon nanowires capable of detecting VOCs. Methods of use in the diagnosis of diseases including various types of cancer are disclosed.

[0015] Notwithstanding the above, screening and detection methods and devices known in the art are expensive and complicated, as they are not designed for ease of use. Moreover, individuals find it difficult to undergo early cancer screening due to the associated costs and complexity. In 2017, the World Health Organization published the "WHO Position paper on mammography screening," stating that new solutions for breast cancer screening are urgently needed. Furthermore, breast cancer has been identified by the European Commission as one of the priorities in research. Thus, there is a need for non-invasive, inexpensive, sensitive, and at-home disease (e.g., cancer) screening solutions. Summary of the Invention

[0016] It is an object of the present invention to provide a non-invasive and non-irradiating disease screening solution that improves upon screening methods and devices known in the art.

[0017] The present invention is based on artificial intelligence (AI) applied to the analysis of human samples, which provides a low-cost, easy-to-use, non-invasive and permanent cancer screening solution that can be used in multiple settings, such as in the user's home environment or in a clinical environment. Furthermore, the present invention can be performed in many ways, for example as a specific self-screening without the need for hospitals or trained personnel, or by a medical professional, such as a nurse or gynecologist, for a painless and radiation-free rapid test. Furthermore, since the built-in classification algorithm relies on AI and / or machine learning (ML), the more samples analyzed, the higher the classification rate and therefore the earlier the disease can be predicted.

[0018] Unlike mammography, the present invention can diagnose cancer without increasing the risk of developing cancer. Moreover, it is radiation-free and non-invasive, as cancer can be diagnosed from a routine urine sample collection, providing high sensitivity (few cases left undiagnosed), precision, accuracy, and high reliability (>95% when classifying metastatic advanced stage breast cancer).

[0019] Another advantage over the prior art is that a portable medical screening device is provided that is suitable for home diagnosis, thus reducing the severity of breast cancer in the population through early detection of breast cancer. In order to solve the above-mentioned problems, the object of the present invention is to provide a portable breast cancer diagnostic device that can be easily carried by a user anytime and anywhere and that can diagnose breast cancer by simple urine collection. In particular, the portable breast cancer diagnostic device is potentially effective in early screening of breast cancer.

[0020] A user can exchange data with the portable medical screening device through a user interface such as a mobile application, making the portable medical screening device easy to use and convenient without the need for a medical or bioengineering background. The device can be used by different users interchangeably and indefinitely over time, making it a reusable medical device. Finally, the device is low cost compared to current screening methods with very high medical costs and can be used many times by many users.

[0021] The invention is defined by the accompanying independent claims. Embodiments are set out in the dependent claims.

[0022] To that end, the present invention provides, according to a first aspect, a system for screening for disease, in particular cancer such as breast cancer, in a subject, the system comprising a portable medical screening device and artificial intelligence based classification software.

[0023] The portable medical screening device has a collection chamber for collecting a test sample from a subject, such as, but not limited to, urine, blood, sweat, vaginal secretions, exhaled breath, saliva, tears, mucus, feces, earwax, pus, farts, sebum, etc.; a set of chemically sensitive sensors; an analysis chamber configured to assign the set of chemically sensitive sensors, the set of chemically sensitive sensors adapted to detect volatile organic compounds (VOCs) in the test sample and generate an output signal indicative of the presence or absence of VOCs in the test sample as a result of the detection; and a first processing unit operably connected to the set of chemically sensitive sensors to receive the generated output signal.

[0024] The artificial intelligence based classification software may be executed by the first processing unit, by a software application installed on a computing device (e.g., an internet-based device such as a smartphone, tablet or PC, among others), or by a remote processing unit (e.g., a cloud-based server) and is configured to determine an outcome regarding the disease by processing and classifying the generated output signal.

[0025] In the case where the artificial intelligence based classification software is executed by a software application or by a remote processing unit, the portable medical screening device further comprises a communication unit operatively connected to the first processing unit and configured to establish communication with the remote processing unit and / or the software application, e.g., via Bluetooth, WIFI, wired connection, etc. Thus, the portable medical screening device can be connected to a remote processing unit or to another device having the software application. The remote processing unit can be accessed via the aforementioned computing device on which the software application is installed.

[0026] The present invention also provides, according to a second aspect, a method for screening for disease, particularly cancer, in a subject, comprising: collecting at least one test sample of the subject by a collection chamber of a portable medical screening device, detecting VOCs in the test sample by a set of chemically sensitive sensors arranged in an analysis chamber of the portable medical screening device and generating an output signal indicative of an intensity level (concentration) of the VOCs in the test sample as a result of the detection, receiving the generated output signal by a first processing unit of the portable medical screening device, and determining a result regarding the disease by processing and classifying the generated output signal by an artificial intelligence based classification software executed by either the first processing unit, a software application installed on a computing device, or a remote processing unit, where if the artificial intelligence based classification software is executed by the software application or the remote processing unit, the software application or the remote processing unit is operatively connected to the communication unit of the portable medical screening device.

[0027] The present invention also provides, according to a third aspect, a portable medical screening device comprising: a collection chamber for collecting at least one test sample of a subject; a set of chemically sensitive sensors; an analysis chamber configured to assign the set of chemically sensitive sensors, the set of chemically sensitive sensors being adapted to detect VOCs in the test sample and to generate an output signal indicative of the presence or absence of the VOCs in the test sample as a result of the detection; and a first processing unit operably connected to the set of chemically sensitive sensors to receive the generated output signal.

[0028] The terms "subject," "user," and "patient" are used interchangeably herein and refer to any mammalian subject, particularly a human subject, whose samples are screened by the systems, devices, and methods of the present invention.

[0029] According to the present invention, the first processing unit (or computational component) can process the information on-edge (i.e., on the portable medical screening device itself) or pass the information (i.e., the generated output signal) to a communication unit, which transmits the information to a remote processing unit or a software application, which can also be connected to a remote processing unit. Thus, the information can be classified on the disclosed portable medical screening device itself, or on the software application, or on the remote processing unit. The results of the data classification can be displayed, for example, via an output screen on the portable medical screening device, via any user interface on the medical device (color-coded LED system, emitted sounds, etc.), or on a computing device on which the software application is installed. In some embodiments, the first processing unit can be a microprocessor and / or microcontroller, such as an Arduino.

[0030] System Configuration In one embodiment, a system according to a first aspect comprises a portable medical screening device and artificial intelligence (AI) based classification software, the AI ​​based classification software being executed by a first processing unit of the device.

[0031] In one embodiment, the system comprises a portable medical screening device (including the communication unit described above), an AI-based classification software, and a software application installed on a computing device, the AI-based classification software being executed by the software application. The communication unit is operatively connected to a first processing unit of the device and configured to establish communication with the software application.

[0032] In one embodiment, the system comprises a portable medical screening device (including a communication unit), an AI-based classification software, and a remote processing unit, the AI-based classification software being executed by the remote processing unit. The communication unit is operatively connected to a first processing unit of the device and configured to establish communication with the remote processing unit.

[0033] In one embodiment, the system comprises a portable medical screening device (including a communication unit), an AI-based classification software, and a software application installed on a computing device and a remote processing unit, the AI-based classification software being executed by the software application. The communication unit is operatively connected to a first processing unit of the device and configured to establish communication with the remote processing unit and the software application. Similarly, the software application and the remote processing unit are operatively connected and can communicate between them.

[0034] In one embodiment, the system comprises a portable medical screening device (including a communication unit), an AI-based classification software, and a software application installed on a computing device and a remote processing unit, the AI-based classification software being executed by the remote processing unit. The communication unit is operatively connected to a first processing unit of the device and configured to establish communication with the remote processing unit and the software application. Similarly, the software application and the remote processing unit are operatively connected and can communicate between them.

[0035] In some embodiments, the portable medical screening device includes a hardware board consisting of a first processing unit and a communication unit.

[0036] Memory or database, health data and / or demographic data In some embodiments, the system further comprises at least one memory or database configured to store health and / or demographic data of the subject and / or a plurality of healthy and / or unhealthy individuals. The memory or database is used by the AI-based classification software and communicates with either the first processing unit, the remote processing unit, and / or the software application. Any data contained in the memory or database can be used, extracted, and / or updated from the memory or database set without modifying the entire memory or database set. This memory or database allows the data therein to be compared with the subject's own data in order to convert the comparison result into a result.

[0037] In one embodiment, the memory or database is included in the remote processing unit. In another embodiment, the memory or database is remote and associated with the remote processing unit and / or the software application. In another embodiment, the memory or database is included in the software application. In another embodiment, the memory or database is included in the first processing unit.

[0038] In one embodiment, the system is comprised of at least two entities / elements: a portable medical screening device and at least one memory or database.

[0039] In one embodiment, the system is comprised of at least three entities / elements: a portable medical screening device, a software application, and at least one memory or database.

[0040] In one embodiment, the system is composed of at least three entities / elements: a portable medical screening device, a remote processing unit, and at least one memory or database.

[0041] In one embodiment, the system is comprised of at least four entities / elements: a portable medical screening device, a remote processing unit, a software application, and at least one memory or database.

[0042] As explained above, the memory or database can store the health data and / or demographic data of the subject and / or the plurality of healthy and / or unhealthy individuals. In one embodiment, the AI-based classification software is further configured to determine a disease outcome (i.e., a data classification outcome) taking into account at least the stored health data and / or demographic data of the plurality of healthy and / or unhealthy individuals and / or the subject. The memory / database can also store the determined disease outcome to be taken into account in a later screening process.

[0043] The term "health data" in the context of the present invention refers to any information related to the physical or mental health of an individual, and further refers to all types of data related to health status, personal preferences regarding treatment options, health insurance or policy numbers, all kinds of treatment reports, socio-economic parameters related to health and wellness, or historical health background such as illnesses in the past years. Health data refers to, but is not limited to, the medical history of a user or individual, the family medical history of a subject, medication intake, smoking habits or drug consumption.

[0044] The term "demographic data", in the context of the present invention, refers to any information relevant to the study of a population based on factors such as age, race, sex, etc. Demographic data also refers to statistically expressed socio-economic information including employment, education, income, marriage rates, birth rates, mortality rates, etc. Demographic data may refer to, but is not limited to, age, gender, sex, ethnicity, place of birth / residence, or other habits / status such as frequency of sports practice.

[0045] In certain embodiments, the health and / or demographic data of a user / several healthy and / or unhealthy individuals may include age, sex, gender, ethnicity, place of birth / residence, weight, height, body mass index, habits related to drug consumption (e.g., tobacco, alcohol, cannabis and other toxic substances), the user's medical history, family medical history, family cancer history, the user's own cancer history (e.g., breast cancer, lung cancer, prostate cancer, colorectal cancer, melanoma, bladder cancer, kidney cancer, non-Hodgkin's lymphoma), tumor histology, tumor behavior (in situ or invasive), tumor profile (luminal A, luminal B HER2 / neu+ ... HER2 / neu-, HER2 / neu+ or Triple-), tumor size, tumor TNM classification, tumor stage (I, II, III or IV), cancer detection method (screening or clinical), tumor histological grade (G1, G2 or G3), menstrual, menarche and menopausal characteristics (e.g., cycle, timing), number of children born, number of premature births, number of miscarriages, number of living children, reproductive status (e.g., under fertility treatment, pregnant, lactating, postpartum), history of breast disease (e.g., chest pain during menstrual cycle, hormonal changes, swelling, breast lumps or tenderness, pouches, etc.), These may include breast cancer, cysts, fibroadenomas, general lumps, breast discharge, sore / cracked / itchy nipples, inverted nipples), endocrine disorders (e.g., diabetes, thyroid disorders, polycystic ovary syndrome, low testosterone, osteoporosis) or dietary type (e.g., omnivory, semi-vegetarian, "chicken"-arianism, pescatarianism, vegetarian, dairy vegetarian, ovovegetarianism, apivegetarianism, vegan, raw vegetarian, granivorianism, fruitarianism, ketogenic diet).

[0046] In particular, the health and / or demographic data of the user / several healthy and / or unhealthy individuals may include age, gender, habits related to drug consumption, the user's medical history, family medical history, family cancer history, the user's own cancer history, and / or breast disease history.

[0047] The memory or database may contain different types of information according to the screening configuration. For example, the memory or database may only contain health and / or demographic data of a plurality of healthy and / or unhealthy individuals if the subject's data is not yet stored in the memory or database, e.g., it is the first screening of the subject. In another example, the memory or database may contain health and / or demographic data of a plurality of healthy and / or unhealthy individuals as well as health and / or demographic data of the subject, e.g., the subject has already introduced its own data from a previous screening, which data is already stored in the memory or database.

[0048] In certain embodiments, the memory or database includes health and / or demographic data for a plurality of healthy and / or unhealthy individuals.

[0049] In certain embodiments, the memory or database comprises health and / or demographic data of a subject and a plurality of healthy and / or unhealthy individuals.

[0050] In the present invention, it is understood that the data stored in the memory or database can be raw data and / or processed data. In some embodiments, the data contained in the memory or database can be raw data, i.e., data that has not been processed by AI-based classification software. In some embodiments, the data contained in the memory or database can be processed data, such as patterns extracted from some post-screening data.

[0051] In some implementations, the memory or database includes health and / or demographic data of a plurality of healthy and / or unhealthy individuals stored therein, and the AI-based classification software includes determining a disease outcome taking into account the stored health and / or demographic data.

[0052] In some implementations, the method also includes obtaining health and / or demographic data of the subject and storing the obtained data in a memory or database, and the AI-based classification software determines an outcome regarding the disease taking into account the stored health and / or demographic data of the subject.

[0053] In some embodiments, the health data and / or demographic data is obtained via a user interface of a software application or a portable screening medical device.

[0054] Portable Medical Screening Device In the present invention, a portable medical screening device can be adapted to house different elements to perform screening. The portable medical screening device that encloses a set of elements is sized and configured to be easily handled and used by an individual, thus enabling in-home screening.

[0055] In one embodiment, the portable medical screening device comprises a communication unit operably connected to the first processing unit and configured to establish communication with a remote processing unit and / or a software application installed on the computing device.

[0056] In one embodiment, the portable medical screening device includes a grid disposed above the collection chamber and / or a drawer for opening, closing and supporting the collection chamber.

[0057] In some implementations, the portable medical screening device can be comprised of a box adapted to enclose the collection chamber, the grid, the analysis chamber, the first processing unit, and the communication unit, the box having a hollow portion for the introduction of a drawer.

[0058] In particular, the portable medical screening device comprises a grid disposed above the collection chamber, and / or a drawer for opening, closing and supporting the collection chamber, the portable medical screening device being comprised of a box adapted to enclose the collection chamber, the grid, the analysis chamber, the first processing unit and the communication unit, the box having a hollow portion for the introduction of the drawer.

[0059] In one embodiment, the device includes a collection chamber for collecting a human sample, such as a urine sample. The test sample may be collected directly into the device or may be collected in a separate device or container and then introduced into the disclosed device.

[0060] In some embodiments, the drawer can be filled with an insulating material. Additionally, the drawer can include a hole (a hole configured to allocate a collection chamber) whose walls are configured to irradiate heat to the sample where the test sample can be collected. In certain embodiments, the entire wall irradiates heat to the sample via a thermal blanket covering the wall.

[0061] In the present invention, the portable medical screening device is an element used by the subject to perform screening. Furthermore, according to the above-mentioned system configuration, the system can further include a software application installed on a computing device, particularly a mobile phone or tablet, and can be configured to establish communication with the first processing unit of the device. Similarly, the system can further include a remote processing unit. The remote processing unit can be located in the cloud or on a computing device such as a PC, tablet, or laptop, among others. The communication between the different system elements can be wired or wireless (e.g., via Bluetooth, WIFI, satellite, or 5G). As used herein, the term "computing device" refers to any computer hardware having a central processing unit (CPU), including but not limited to a computer, laptop, tablet, personal digital assistant (PDA), smartphone, or smartwatch.

[0062] In one embodiment, the remote processing unit is located, for example, in the cloud, a web browser, a computer, a laptop, a smartphone, or a tablet.

[0063] In one embodiment, the software application is located in, for example, a web browser, a computer, a laptop, a smart phone, a PDA, a tablet, or a smart watch.

[0064] In one embodiment, the software application and / or the remote processing unit is operatively connected wirelessly to the first processing unit, hi a particular embodiment, the wireless connection is via Bluetooth and / or WIFI.

[0065] In certain embodiments, the medical device may be operated in a remote location, in which case the medical device has the ability to connect directly to the cloud via satellite or 5G using a special connectivity module embedded in the device specifically for such purpose. In certain embodiments, the wireless connection is via satellite and / or 5G.

[0066] User Interface In the present invention, the subject's health data and / or demographic data are collected through a user interface, then processed by the AI-based classification software and / or stored in a memory / database. The user interface can be displayed either by a software application or a portable medical screening device. For example, the user interface is displayed by a software application installed on a computing device, and the computing device is a smartphone.

[0067] The software application may be developed using different programming languages, including but not limited to XML, JavaScript, Swift, Python, SQL, PHP, Ruby, C, C++, or C Sharp. The software application may run on different operating systems, including but not limited to Android, iOs, Linux, Mac OS, Windows MS, Ubuntu, Android, Fedora, Solaris, Free BSD, or Chrome OS. In a particular embodiment, the software application runs on an Android smartphone and the software application is developed using XML and / or Java. In a particular embodiment, the software application runs on an iOs smartphone and the software application is developed using Swift.

[0068] Once the software application is installed on the computing device, the subject may register (i.e., create an account) by introducing personal data such as name or nickname, email, birth date, gender, and a password. After logging in, the software application displays different interfaces such as a personal information interface (e.g., the subject's health and demographic data) (Figure 9, panel A), a screening history interface (Figure 9, panels C and D), a screening test execution interface (Figure 10, panels A-F), a configuration / settings interface (Figure 9, panel B), among others.

[0069] The screening test execution interface allows the software application to intuitively guide the subject and / or medical professional through the screening process until a diagnosis is displayed. The subject can decide when to start the screening test execution, and the interface displays instructions to the user, for example, to enable Bluetooth, WIFI, and / or cellular settings (FIG. 10, panel A) and introduce a test sample into the portable medical screening device (FIG. 10, panel C). After these instructions, the data is automatically processed and sent back to the computer device (if the data is processed in a processing unit not located in the computer device), and the computer device displays the results regarding the disease in an understandable manner (FIG. 10, panel F). In some embodiments, the software application can guide the subject through the screening process.

[0070] The software application may further include several options for the user to receive the results regarding the disease, including displaying the results on a computing device, receiving a call from a medical professional (such as a doctor, nurse, or psychologist) or setting up a doctor's appointment, or the diagnosis may arrive to the medical professional as a notification or tumor report.

[0071] The personal information interface allows users to introduce health and / or demographic data that is further considered by the AI-based software to generate accurate results.

[0072] In some embodiments, the subject's health and / or demographic data is collected via a user interface of the software application and then stored in a memory / database.

[0073] Additionally, the software application may incorporate an embedded virtual agent, a chatbot powered by natural language processing (NLP), that continuously performs psychological assessments of the user while delivering results. The virtual agent analyzes the user's text-based responses (e.g., via a chatbot built into the software application) and determines whether the user is processing emotions accordingly or according to avoidance behavior patterns and thus in need of psychological assistance. Alternatively or additionally, the virtual agent may analyze the user's facial expressions (captured by a computer camera) and process them using a convolutional neural network to correlate them with emotions the user may be feeling.

[0074] Depending on the environment in which the screening is performed, different software applications can be used. In one embodiment, if the screening is intended to be performed at home, the software application is directed to the subject and includes, for example, a forum for multiple purposes (e.g., social networking, scientific article sharing, or health data sharing). In another embodiment, the screening is intended to be performed at home by subjects with limited IT capabilities (e.g., elderly people) with a simplified software application and mainly including a result display. In yet another embodiment, the screening is intended to be performed at home, and the software application is directed to the subject and does not include a result display, but rather includes the subject's screening history (results are sent to a medical professional). In another embodiment, the screening is intended to be performed in a clinical environment (e.g., hospital, nursing home, pharmacy, ambulance), and the software application is installed on a PDA or tablet, is managed by a medical professional, and mainly includes a result display.

[0075] Chemically Sensitive Sensors In the present invention, the portable medical screening device includes a set of chemically sensitive sensors for detecting the presence or absence of VOCs in a test sample, with a first processing unit operatively connected to receive the generated output signal. In certain embodiments, the first processing unit is connected to the set of chemically sensitive sensors via a wired connection.

[0076] The present invention is based on the principle that the presence of disease (e.g., breast cancer) is indicated not by specific chemical compounds found in a test sample (e.g., urine), but rather by the proportional relationship between all chemical compounds found in, for example, urine.

[0077] The set of chemically sensitive sensors in the analysis chamber is not specific to a few biomarkers (e.g., breast cancer biomarkers), but rather is sensitive to multiple sample properties or compositions. Furthermore, sample classification does not simply depend on the presence or absence of a particular set of, for example, cancer biomarkers.

[0078] The compounds captured by the chemically sensitive sensors are not direct, e.g. cancer metabolites. Instead, they are residual metabolites resulting from cellular metabolism. For example, the aforementioned metabolites change in concentration due to cancer, as cancer causes changes in the body's physiology. This strategy of targeting ratios between chemicals instead of specific biomarkers is inspired after the olfactory cortex of dogs.

[0079] According to the present invention, the generated output signal is indicative of the presence or absence of a VOC in the test sample. In one embodiment, the generated output signal comprises an electrical signal whose voltage is proportional to the concentration of the VOC. For example, the output signal is "0" when no VOC is present, "1023" when the test sample is saturated with the VOC, and the signal has an intermediate value when an amount of the VOC is present. That is, the output signal is in particular an analog signal.

[0080] In one embodiment, the set of chemically sensitive sensors includes a photoionization detector (PID), a flame ionization detector (FID), a metal oxide sensor (MOS) and / or a silicon photonic ring resonator (SPRR), or any type of gas sensing structure consisting of a self-assembled monolayer coated sensor array that is sensitive to volatile organic compounds and has a concentration-independent fingerprint. In one embodiment, the set of chemically sensitive sensors includes a metal oxide sensor and / or an organic compound sensitive sensor (e.g., a semiconductor-based chemiresistive sensor, an electrochemical sensor, a metal oxide sensor and a PID sensor). In a particular embodiment, the MOS is a tin oxide sensor.

[0081] These sensors can be used in combination with or without nanotechnology or microfluidic technology to have a human body fluid circulated through a microfluidic circuit before or during analysis, measure its ability to flow through the microfluidic circuit, and also measure its odor. In one embodiment, a set of chemically sensitive sensors is used in combination with microfluidic technology. In one embodiment, a set of chemically sensitive sensors is used in combination with nanotechnology technology.

[0082] In one embodiment, the portable medical screening device comprises at least one chemically sensitive sensor. In particular, the portable medical screening device comprises 1-20 chemically sensitive sensors. More particularly, the portable medical screening device comprises 5-10 chemically sensitive sensors. More particularly, the portable medical screening device comprises 9 chemically sensitive sensors.

[0083] The set of chemically sensitive sensors can be configured to detect VOCs all at the same time. In other embodiments, each of the chemically sensitive sensors can be configured to detect VOCs at a given time interval. Similarly, in some embodiments, each of the chemically sensitive sensors can be responsive to one or more VOCs.

[0084] In one embodiment, the set of chemically sensitive sensors detects VOCs over a range of minutes, preferably less than 45 minutes. In a particular embodiment, the set of chemically sensitive sensors detects VOCs over a time interval of 10-120 minutes, preferably 25-35 minutes.

[0085] Test sample In one embodiment, aside from detecting the presence or absence of VOCs, the portable medical screening device further detects several properties of the test sample, such as its chemical composition, odor, temperature, viscosity (i.e., Reynolds number), density, color, conductivity, capacitance, boiling time, its weight or translucency, refractive index, change in any chemical / physical property over time or while the sample is heated, among other physical / chemical properties.

[0086] In some embodiments, it may be necessary to apply some chemical or physical stimulus to the sample, such as, but not limited to, a chemical reagent or any other substance, physical vibration such as sonication, heat, pressure, movement such as decanting the substrate, and other stimuli. Additionally, observations may be made based on a change in said observation or the rate at which the change occurs.

[0087] In another embodiment, the portable medical screening device generates a particular response (ie, pattern) taking into account the properties of the test sample, or taking into account the particular odor of the test sample.

[0088] The present invention is capable of analyzing at least one test sample and generating an output signal indicative of the presence or absence of a VOC.

[0089] The test sample may be, but is not limited to, urine, blood, sweat, tears, vaginal mucus, vaginal secretions, menstrual blood, menstrual mucus, breath, saliva, serous fluid, feces, sperm, tears, mucus, stool, earwax, pus, farts, and sebum. In particular, the test sample may be urine, breath, vaginal secretions, menstrual blood, feces, and saliva. More particularly, the test sample is urine.

[0090] In some embodiments, the systems, devices, and methods collect and analyze two test samples, e.g., urine and breath, urine and menstrual blood, or breath and menstrual blood. In some embodiments, the systems, devices, and methods collect and analyze three test samples, e.g., urine, breath, and menstrual blood.

[0091] In some embodiments, the drawer is filled with an insulating material and comprises a hole (a hole configured to allocate a collection chamber), so that the collection chamber containing the sample can be introduced into the hole and better retained. Furthermore, the wall of the hole can further irradiate heat to the sample. In some embodiments, this is done by the inside of the hole wall being covered with a thermal blanket (i.e. a dissipative resistor that converts electrical energy into heat). The thermal blanket can thus heat the wall of the hole that is in direct contact with the sample or the collection chamber, and thus transfer heat to the collection chamber and the sample. To control the sample heating process, in one embodiment, the first processing unit is operatively connected to a relay that can mechanically start and stop the system from supplying voltage to the temperature sensor that is in contact with the thermal blanket and the hole wall. The temperature sensor may be integrated in the thermal blanket, may be placed directly in contact with the hole wall, may be placed directly in contact with the collection chamber, or may be placed directly in contact with the sample. The temperature sensor is operatively connected to the first processing unit, which continuously reports the temperature (or an approximation) at which the sample is present at each time point. Based on this information, the first processing unit is configured to make a decision to keep the thermal blanket on or off, thus controlling the temperature of the specimen as well as the rate at which the specimen is heated. In doing so, the thermal blanket, the temperature sensor, the relay, and the first processing unit form a feedback control loop to set the temperature of the specimen.

[0092] In one embodiment, the set of chemically sensitive sensors is configured to detect VOCs in the test sample when the temperature of the test sample reaches a temperature of at least 25° C. In particular, the test sample reaches a temperature of at least 40° C. In particular, the test sample reaches a temperature of at least 50° C. In yet another particular embodiment, the test sample reaches a temperature of 25° C. to 90° C. In yet another particular embodiment, the test sample reaches a temperature of 40° C. to 90° C. The accuracy and optimization of the screening is improved when the test sample reaches such temperatures. The chemically sensitive sensors stop detecting VOCs when the temperature in the test sample reaches a temperature of, for example, 80 to 95° C. depending on the temperature selected to heat the sample.

[0093] The test sample may be collected directly into the portable medical screening device or may be collected in a separate device or container and then introduced into the disclosed device. In certain embodiments, the container is a collection chamber.

[0094] Artificial Intelligence-Based Classification Software The artificial intelligence based classification software may include one or more machine learning / artificial intelligence based (ML / AI) or statistical classification algorithms, including, but not limited to, artificial regression neural networks (RNNs) such as long short-term memory (LSTM), artificial neural networks (ANNs), convolutional neural networks (cNNs or ConvNets), principal component analysis (PCA), multi-layer perceptrons (MLPs), generalized regression neural networks (GRNNs), fuzzy inference systems (FISs), self-organizing maps (SOMs), radial bias functions (RBFs), genetic algorithms (GASs), neuro-fuzzy systems (NFSs), adaptive resonance theory (ARTs), partial least squares (PLSs), multiple linear regression (MLRs), principal component regression (PCRs), discriminant function analysis (DFAs), linear discriminant analysis (LDAs), cluster analysis, and nearest neighbors. In particular, the AI / ML classification algorithms are RNNs such as LSTMs and / or ANNs.

[0095] Additional algorithms suitable for identifying patterns of VOCs and quantifying their concentrations can include, but are not limited to, Fisher Linear Discriminant Analysis (FLDA), Soft Independent Modeling of Class Analogies (SIMCA), k-Nearest Neighbors (kNN), and fuzzy logic algorithms. In some embodiments, FLDA and Canonical Discriminant Analysis (CDA) and combinations thereof are used to compare the output signature with available data from a database.

[0096] Many of the algorithms are neural network-based algorithms. A neural network has an input layer, a processing layer, and an output layer. Data extracted by a sensor from a sample is fed as input to the input layer and flows through the processing layers until it reaches the output layer. At each processing layer, the information is modified and passed to the next layer. The information that reaches the output layer leads to a classification result. The processing layer is composed of nodes that simulate neurons by their interconnections to the nodes. In operation, when a neural network is combined with a sensor array, the sensor data is propagated through the network. In this way, a series of vector-matrix multiplications are performed, and unknown analytes can be easily identified and determined. The neural network is trained by correcting erroneous or undesired outputs from a given input. Similar to statistical analysis, which reveals underlying patterns in a collection of data, neural networks find consistent patterns in a collection of data based on predefined criteria.

[0097] Disease results In the present invention, once the AI-based classification software processes all the inputs (e.g., data acquired by the sensors and user data introduced by the software application), the software application or the portable medical screening device can display the results to the user.

[0098] The term "result" as used herein refers to the determination or detection of a disease (e.g., a diagnosis of a disease), which may include, among other things, an indication as to the presence or absence of a disease, a probability of risk of having a disease, an indication of the extent and / or type (or subtype) of disease and / or size of a tumor, a recommendation for a particular treatment, a prediction of the progression of a particular disease.

[0099] In some embodiments, if the sample classification is performed by AI-based classification software hosted on a software application, the results are shown on the computer device. In another embodiment, if the sample classification is performed by AI-based classification software hosted on a software application, the results are sent via the Internet, Bluetooth, or another means and shown on the portable screening device. In another embodiment, if the sample classification is performed by AI-based classification software hosted on a screening device or a cloud-based server, the results can be sent, for example, via the Internet or Bluetooth, from such screening device or from the cloud server to the software application, where the results are displayed. In another embodiment, if the sample classification is performed by AI-based classification software hosted on a cloud-based server, the results can be sent, for example, via the Internet or Bluetooth, from such screening device to the portable screening device, where the results are displayed. In yet another embodiment, if the sample classification is performed by AI-based classification software hosted on the first processing unit, the results are shown on the screening device. This allows the user to check the results as soon as the classification is completed.

[0100] In one embodiment, the AI-based classification software is configured to determine a disease outcome by processing and classifying the generated output signal. In some embodiments, the disease outcome includes an indication of the presence or absence of a disease. For example, the outcome may be in the form of a report or a binary outcome (e.g., values ​​"0" or "1" for "cancer" and "healthy", etc.). The AI-based classification software may determine an outcome through a risk probability of suffering from a disease. In some embodiments, the disease outcome is a risk probability of suffering from a disease.

[0101] Furthermore, the artificial intelligence is also implemented to provide information about the test sample being analyzed, i.e., tumor size (if applicable), type of cancer (if applicable), and other information. In certain embodiments, the results include an indication of the extent and / or type (or subtype) of the disease, and / or an indication of the size of the tumor if the disease is cancer. The term "disease" can also include any condition, disorder, syndrome, or state of a subject that is altered compared to a healthy subject, and refers to a specific abnormal condition that adversely affects the structure or function of all or part of an organism. A subject's condition or altered state is often associated with specific signs and symptoms that may cause a disease, disorder, syndrome, or severe condition.

[0102] Many diseases can be diagnosed, predicted or monitored by the present invention, including, but not limited to, cancer, diabetes, stress, epilepsy, Alzheimer's disease, oral infections, periodontal disease, halitosis, ketosis, yeast infections, urinary tract infections, pneumonia, lung infections, sexually transmitted diseases, vaginitis, nephritis, bilirubin production, kidney disease, uremia, trimethylaminuria, cardiovascular disease, hypercholesterolemia, and gastrointestinal infections.

[0103] The present invention may further be useful in diagnosing other medical disorders including, but not limited to, acute asthma, hepatic coma, rheumatoid arthritis, schizophrenia, ketosis, cardiopulmonary disease, uremia, diabetes mellitus, taste / smell disorders, cystinuria, cirrhosis, histidinemia, tyrosinemia, halitosis, and phenylketonuria.

[0104] Non-limiting examples of cancers that can be detected by the present invention are brain cancer, ovarian cancer, pancreatic cancer, liver cancer, colon cancer, prostate cancer, kidney cancer, bladder cancer, breast cancer, lung cancer, oral cancer, and skin cancer.Specific examples of cancers are: adenocarcinoma, adrenal tumor, ameloblastoma, anaplastic tumor, anaplastic carcinoma of thyroid cells, angiofibroma, hemangioma, angiosarcoma, apdoma, argyrrhoma, arrayoblastoma, ascites tumor cells, ascites tumor, astroblastoma, astrocytoma, ataxia telangiectasia, atrial myxoma, basal cell carcinoma, benign tumor, bone cancer, bone tumor, brain stem glioma, brain tumor, breast cancer, vaginal tumor, Burkitt's lymphoma, carcinoma, cerebellar astrocytoma, cervical cancer, senile hemangioma, cholangiocarcinoma, cholangiocarcinoma, chondroblastoma, chondroma, chondrosarcoma, chorioblastoma, choriocarcinoma, laryngeal cancer, colon cancer, general Acute lymphoblastic leukemia, craniopharyngeal carcinoma, cystic carcinoma, cystoma, fibroblastoma, fibroblastoma, intraductal carcinoma in situ, ductal papilloma, dysgerminoma, brain tumor, endometrial carcinoma, endothelioma, ependymoma, epithelioma, erythroleukemia, Ewing's sarcoma, extranodal lymphoma, feline sarcoma, fibroadenoma, fibrosarcoma, follicular carcinoma of the thyroid, ganglioneuroma, gastrinoma, glioblastoma multiforme, glioma, gonadoblastoma, hemangioblastoma, hemangioendothelioma, hemangiopericytoma, hematopoietic tumor, hematopoietic tumor, hairy cell leukemia, hamartoma, hepatocellular carcinoma, hepatocellular carcinoma, hepatocellular carcinoma, hepatoma, histoma, hodgkin's tumor, thyroid cancer ... Kinesiosis, hypernephroma, invasive carcinoma, invasive ductal cell carcinoma, insulinoma, juvenile fibroma, Kaposi's sarcoma, renal tumor, large cell lymphoma, acute leukemia, leukemia, hepatoma, liver metastasis, Lucke's carcinoma, lymphadenoma, lymphangioma, lymphocytic leukemia, lymphocytic lymphoma, lymphocytoma, lymphedema, lymphoma, lung cancer, malignant mesothelioma, malignant teratoma, mast cell tumor, medulloblastoma, melanoma, meningioma, mesothelioma, metastatic carcinoma, Morton's neuroma, multiple myeloma, myeloblastoma, myelogenous leukemia, myelolipoma, myeloma, myeloma, myeloblastoma, myeloma, nasopharyngeal carcinoma, nephroblastoma, neuroblastoma, neurofibroma, neurofibromatosis, neuro Glioma, neuroma, non-Hodgkin's lymphoma, oligodendroglioma, optic nerve glioma, osteochondroma, osteogenic sarcoma, osteosarcoma, ovarian cancer, Paget's disease of the nipple, Pancoast tumor, pancreatic cancer, pheochromocytoma, pheochromocytoma, pheochromocytoma, plasmacytoma, primary brain tumor, progestoma, prolactinoma, renal cell carcinoma, retinoblastoma, rhabdomyosarcoma, rhabdomyosarcoma, solid tumor, sarcoma, secondary tumor, seminoma, skin cancer, small cell carcinoma, squamous cell carcinoma, strawberry hemangioma, T-cell lymphoma, teratoma, testicular cancer, thymoma, trophoblastic tumor, neoplasia, vestibular schwannoma, Wilms' tumor, or a combination thereof.

[0105] In some embodiments, the disease is selected from the group consisting of cancer, diabetes, and stress. In particular embodiments, the disease is cancer. In particular embodiments, the cancer can be selected from the group consisting of breast cancer, pancreatic cancer, liver cancer, ovarian cancer, colon cancer, brain cancer, uterine cancer, or lung cancer. In more particular embodiments, the cancer is breast cancer.

[0106] The present invention also relates to the identification of patterns involved in cancer. In particular, the present invention can identify different grades of cancer within and between its stages. Broadly defined, these stages can be considered as non-malignant versus malignant, but also normal versus atypical (optionally including reactive and pre-neoplastic) versus cancerous. Another definition of the stages is normal versus pre-cancerous (e.g., atypical ductal hyperplasia (ADH) or atypical lobular hyperplasia (ALH)) versus cancerous (e.g., carcinoma in situ, such as DCIS and / or LCIS) versus invasive (e.g., carcinoma, such as IDC and / or ILC). The present invention can also be applied to distinguish between normal and non-normal cells, including cancerous and other non-normal cells.

[0107] For example, breast cancer grading is typically performed for cases of invasive ductal carcinoma (IDC) and can be performed for invasive lobular carcinoma (ILC) as well, using cytological criteria such as Nottingham BSR, nuclear morphology, histology, proliferation index (such as assays for PCNA or Ki67), and degree of differentiation to assign a grade of I, II, or III to a particular breast cancer sample. Grade I is typically when the cells are still well differentiated and are usually positive for estrogen receptor (ER). Grade III is typically when the cells are poorly differentiated and are usually negative for ER. Grade II is typically when the cells have characteristics intermediate between grade I and grade III and may constitute approximately 60% of all samples measured.

[0108] For example, the "stage"(s) (or equivalents) of breast cancer refer to the physiological state of breast cells, e.g., as defined by known cytological or histological procedures (including immunohistology, histochemistry and immunohistochemistry). Non-limiting examples include normal versus abnormal, noncancerous versus cancerous, the different stages described herein (e.g., hyperplasia, carcinoma, and invasive), and grades within the different stages (e.g., grades I, II, or III or their equivalents within the cancerous stages).

[0109] In one embodiment, the disease outcome includes an indication of the extent of the disease, such as disease stage, histological grade, tissue type, behavior, profile, TNM staging (tumor / node / metastasis), etc. In certain embodiments, the disease stage may be I, II, III, or IV. In certain embodiments, the disease behavior may be hyperplastic, carcinoma, or invasive. In certain embodiments, the disease histological grade may be grade I, grade II, or grade III. In certain embodiments, the disease profile may be Luminal A, Luminal B HER2 / neu+, Luminal B HER2 / neu-, HER2 / neu+, or triple-.

[0110] A subject can use the present invention to diagnose several diseases in one screening. Thus, in one embodiment, the result regarding a disease includes at least one disease. In a particular embodiment, the result includes an indication of the type of disease. In a particular embodiment, breast cancer, pancreatic cancer, lung cancer, colon cancer, ovarian cancer, liver cancer and uterine cancer can be diagnosed. In a particular embodiment, breast cancer, pancreatic cancer and lung cancer can be diagnosed.

[0111] Tumor volume is important for treatment planning and for assessing the extent of tumor regression and disease prognosis. Thus, the present invention can also determine the size and volume of the tumor itself. In one embodiment, the disease-related outcome includes an indication of tumor size. In a particular embodiment, the disease is cancer and the outcome includes an indication of tumor size.

[0112] Additionally, it may include treatment recommendations based on observations of past treatments and their success rates. In one embodiment, the disease-related results include treatment recommendations.

[0113] In some embodiments, the disease outcome is a prediction of disease progression based on a comparison between the current sample and a previously analyzed sample.

[0114] The present invention can be used for disease screening and monitoring purposes.For example, patients diagnosed with breast cancer can be screened periodically to obtain diagnosis of disease stage / grade at different time points.In some embodiments, the present invention can be used to monitor disease progression, regression, recurrence, and / or response to treatment in subjects.

[0115] In certain embodiments, the present invention is further used to monitor the recurrence of a tumor in a subject. In one embodiment, the present invention provides a clinical algorithm for individualized monitoring of the course of a tumor and its treatment in a given patient. Monitoring can be pre-treatment, during treatment, or post-treatment (e.g., after surgery, radiation, and / or chemotherapy). Thus, the systems and methods of the present invention can be used to monitor tumor progression, spread, treatment, metastasis, and / or recurrence.

[0116] In one embodiment, the disease result has an accuracy (i.e., classification rate) of at least 75%. In certain embodiments, the disease result has an accuracy of 75%, 76%, 77%, 78%, 79%, 80%, 81%, 82%, 83%, 84%, 85%, 86%, 87%, 88%, 89%, 90%, 91%, 92%, 93%, 94%, 95%, 96%, 97%, 98%, 99% or 100%. In certain embodiments, the disease result has an accuracy of at least 95%. In certain embodiments, the disease result has an accuracy of 95%, 96%, 97%, 98%, 99% or 100%. In another embodiment, the accuracy % is 100%.

[0117] In one embodiment, the result for the disease has a sensitivity of at least 75%. In certain embodiments, the result for the disease has a sensitivity of 75%, 76%, 77%, 78%, 79%, 80%, 81%, 82%, 83%, 84%, 85%, 86%, 87%, 88%, 89%, 90%, 91%, 92%, 93%, 94%, 95%, 96%, 97%, 98%, 99% or 100%. In certain embodiments, the result for the disease has a sensitivity of at least 95%. In certain embodiments, the result for the disease has a sensitivity of 95%, 96%, 97%, 98%, 99% or 100%. In another embodiment, the sensitivity % is 100%.

[0118] In one embodiment, the result for the disease has a specificity of at least 75%. In certain embodiments, the result for the disease has a specificity of 75%, 76%, 77%, 78%, 79%, 80%, 81%, 82%, 83%, 84%, 85%, 86%, 87%, 88%, 89%, 90%, 91%, 92%, 93%, 94%, 95%, 96%, 97%, 98%, 99% or 100%. In certain embodiments, the result for the disease has a specificity of at least 95%. In certain embodiments, the result for the disease has a specificity of 95%, 96%, 97%, 98%, 99% or 100%. In another embodiment, the specificity rate is 100%.

[0119] Screening Method Embodiments The step of detecting VOCs in a test sample refers to determining the presence or absence of VOCs depending on the concentration of the VOCs, i.e. the intensity of some VOCs present in the test sample, and the proportionality between their intensities. Each sensor of a set of chemically sensitive sensors outputs an electrical signal whose intensity is proportional to the concentration of various VOCs in the sample. Each sensor has a different affinity for a range of VOCs. Some VOCs can be detected by more than one sensor. The combined response of all sensors therefore provides information about the composition of the sample. The concentration of VOCs is analyzed at each time point of the detection period, which can last for example 30 minutes while the sample is heated. The sample headspace is the inside of the sample container (or inside the drawer if no sample container is used), i.e. all the space of the analysis chamber that is not occupied by the sample. Before starting the analysis, the sample headspace is filled with ambient air at room temperature. During the analysis, the thermal blanket transfers heat to the sample container and / or the sample, while the VOCs contained in the sample evaporate at their specific evaporation temperature. Thus, the sample and sample headspace change in composition over time as VOCs evaporate from the sample and become present in the sample headspace. Thus, the sensor response changes over time. Typically, VOCs, which are composed of smaller volatile molecules, evaporate first. An electrical output signal is generated from the information extracted during the sensing period, which indicates the concentration of VOCs in the test sample as a result of the detection. For example, the output signal is "0" when no VOCs are present, "1023" when the test sample is saturated with VOCs, and the signal takes on intermediate values ​​when a certain amount of VOCs is present. That is, the electrical output signal is specifically an analog signal.

[0120] The generated electrical output signal is then transmitted to the first processing unit via a wired or wireless connection. Once the first processing unit receives the VOC data extracted from the test sample, this information needs to be processed to determine a disease outcome by processing and classifying the generated output signal via AI-based classification software.

[0121] Before utilizing the AI-based classification software, the system checks for the latest version of the classification software in the version bucket. Before any classification, the system ensures that the latest version of the classification algorithm is installed. The AI-based classification software periodically receives real-time updates.

[0122] The AI-based classification software includes different algorithms that use different data to generate results. For example, the algorithm is responsible for processing the subject's health and demographic data obtained from a software application installed on a computing device. The algorithm is responsible for processing the data obtained from a set of chemically sensitive sensors, where the algorithm merges all the input data and recognizes patterns in the information stored in a memory or database.

[0123] In one embodiment, the AI-based classification software includes at least one ML / AI-based classification algorithm. In certain embodiments, the AI-based classification software includes an ANN that takes as input the subject's health and demographic data. In certain embodiments, the AI-based classification software includes an LSTM network that receives the sensor data as input.

[0124] The input consists, for example, of a 2D data array. Each row of the array is a 1D vector containing the response of one sensor. Every element of each column relates to an instant in time. The data array therefore has 9 rows (one per sensor) and as many columns as the number of times the sample was odorized, which can be, for example, 100 columns.

[0125] In the present invention, the determined result can be stored in a memory or database. In this case, the AI-based classification software can also consider the first or initial screening result of the subject in a subsequent screening operation to determine its further screening result. Thus, in one embodiment, the method further comprises storing the determined result regarding the disease in a memory or database, and the artificial intelligence-based classification software comprises determining the result regarding the disease by considering the stored result.

[0126] With each new analysis, the classification algorithm reaches an output that is determined by whether the input signal is more similar to the previously analyzed diseased or control (healthy) samples. This is possible because the AI ​​algorithm in the classification software has been previously trained, i.e. the weights of all neurons in the network have been adjusted to minimize the discrimination error, i.e. the difference between the predicted probability of having cancer and the absolute truth (0% or 100% for all samples, as determined by a biopsy performed after the portable medical screening device analysis).

[0127] In addition, instead of generating knowledge about test samples uniquely based on their composition, the present invention proposes in some embodiments that each time a new test sample is analyzed and some data is acquired from the human sample by the screening device, the acquired data is compared with all data in the database, i.e. sensor data from all previously analyzed samples, patient data previously entered by the user via the user interface, and / or sensor / patient data from other users. The classification of the new test sample is evaluated based in particular on the extent to which it resembles previously analyzed data. The class depends on the extent to which the new test sample resembles previously analyzed cancerous samples and the extent to which the new test sample resembles previously analyzed control (healthy) samples. That is, machine learning is performed to classify the test sample.

[0128] Thus, the present invention provides a portable screening solution that can be used to diagnose cancer using different types of test samples from a subject. The disclosed device can be used multiple times by one subject or multiple subjects.

[0129] The foregoing and other advantages and features will be more fully understood from the following detailed description of examples and embodiments, taken in conjunction with the accompanying drawings, which should be considered in an illustrative and non-limiting manner. [Brief description of the drawings]

[0130] [Figure 1] 1 illustrates an example of a portable medical screening device according to one embodiment. [Figure 2A] 1 illustrates an example of the internal structure of a portable medical screening device according to one embodiment. [Figure 2B] 1 illustrates an example of placement of a chemically sensitive sensor inside the internal structure of a portable medical screening device, according to one embodiment. [Figure 2C] 1 illustrates an example of placement of a chemically sensitive sensor inside the internal structure of a portable medical screening device, according to one embodiment. [Diagram 3] 1 illustrates a system for screening for disease in a subject, according to one embodiment. [Figure 4] 1 illustrates different examples in which AI-based classification software can be implemented. [Diagram 5] 2 shows another example of the proposed system. [Figure 6-1] Panel A shows a score plot with control (crosses) and breast cancer (triangles) human urine samples projected onto a virtual plane that maximizes the inter-class sample variability. The virtual plane is defined by two vectors, principal components (PC) 1 and PC2. These two vectors represent a linear combination of different features. The samples were analyzed using a gas chromatograph mass spectrometer. [Figure 6-2]Panel B of Figure 6 shows a zoom in to the area where most of the samples are located. Quadrant I (top left) contains 60 samples, of which 49 are controls (81.7%). Quadrants II and III (top right and bottom right, respectively) contain 12 breast cancer samples out of a total of 14 (85.7%). Quadrant IV (bottom left) shows 16 breast cancer samples. [Figure 7] Panel A shows the distribution of control and breast cancer human urine samples after sniffing with the device. These samples were analyzed with the portable medical screening device of the present invention. A virtual plane is defined by two vectors PC1 and PC3. The samples are projected relative to their highest variability plane. Panel B of FIG. 7 shows a zoom into the region of interest. A tendency for breast cancer samples to cluster under the virtual grey dotted line is observed. The control samples show a more dispersed and more variability pattern. [Figure 8] Panels A and B show the accuracy and loss, respectively, by the number of samples used to train the CNN model. Accuracy refers to the proportion of samples that are properly classified out of all collected samples. The loss function is a function that calculates the distance between the current output of the algorithm and the expected output. It therefore describes how well the classification algorithm classifies the samples. As more samples are used to train the model, the accuracy increases and the loss decreases. Samples enter the model one batch at a time, and the model predicts their class. The loss function is used to calculate how far the prediction is from reality. The model is then adjusted in proportion to the loss value. As a result, when the next batch of samples enters the model, its predictions will be more accurate (hence the accuracy will be higher) and the loss value will be lower. These graphs show the change in these values ​​as all batches enter the model. [Figure 9-1] 1 illustrates different interfaces of a software application according to one embodiment: A) A personal information interface (e.g., a subject's health and demographic data); B) A configuration / settings interface. [Figure 9-2]13A-13D show different interfaces of the software application according to one embodiment. C) and D) show the screening history interface. [Figure 10-1] 1 shows a test interface that guides a user through the different steps to perform a screening, including: A) connecting the portable medical screening device to, for example, a computing device, and B) collecting a test sample. [Figure 10-2] 1 shows a test interface that guides a user through the different steps to perform a screening, including: C) introducing the test sample; D) closing the drawer. [Figure 10-3] 1 shows a test interface that guides a user through the different steps to perform a screening: E) starting the screening process, F) displaying the screening results and recommendations according to the results. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0131] The present invention provides systems, devices and methods for screening for disease, particularly cancer, such as breast cancer, in a subject.

[0132] With reference to Figures 1 and 2A-2C, an example of a proposed portable medical screening device (or simply screening device) 100 is shown. According to this embodiment, the screening device 100 consists of a box 101 having a hollow portion for introducing a drawer 102. A user / subject can open the drawer 102 and introduce a test sample to be screened / analyzed into a collection chamber 104 of the screening device 100. A user can open or close the drawer 102 by pulling or pushing a knob or other such means attached to the drawer 102. It should be noted that in other embodiments not shown, the screening device 100 may not include a drawer 102.

[0133] FIG. 2A shows an example of an internal structure 103 of a screening device 100, including a collection chamber 104, a grid 105 (not limiting, since in other embodiments the screening device may not have this component), and an analysis chamber 107. The screening device 100 also includes a set of chemically sensitive sensors 201-209, a computational component (or a first processing unit), and optionally a communication unit, as shown in FIG. 2B and FIG. 2C as a specific embodiment of the present invention. In other embodiments, the set of chemically sensitive sensors may differ in number and type of sensors (particularly, may include 5-20). The internal structure 103 is housed inside a box 101.

[0134] When a test sample (e.g. a human sample such as urine, blood, sweat, vaginal secretions, among others) is introduced into the collection chamber 104, chemical constituents evaporate from the sample, cross the grid 105 (if present) and reach the upper compartment in which the analysis chamber 107 is located, where a set of chemically sensitive sensors 201-209, in particular metal oxide sensors (e.g. tin oxide sensors, among others) and / or organic compound sensitive sensors, can detect VOCs in the test sample and generate an output signal indicative of the presence or absence of such VOCs in the test sample as a result of the detection. In particular, the generated output signal is an (analog) electrical signal whose voltage is proportional to the VOC concentration.

[0135] In particular, the first processing unit and the communication unit (if present) are located at the bottom of the screening apparatus 100, within the internal structure 103 and behind the collection chamber 104. The first processing unit is operatively connected to the set of chemically sensitive sensors 201-209 via wired or wireless connections to receive the generated output signals. The first processing unit in some embodiments may be a microprocessor.

[0136] Although not shown, in some embodiments, the drawer 102 (if present) can be filled with an insulating material and include a hole for inserting a collection chamber containing therein the test sample to be screened / analyzed. Additionally, the inner wall of the hole can be further coated with a thermal blanket. To control the sample heating process, the first processing unit can be operatively connected to a relay that can mechanically start and stop the device from supplying voltage to a temperature sensor in contact with the thermal blanket and the hole wall.

[0137] Figure 3 shows an example of different elements / entities that may be included in the proposed system for screening for diseases, particularly cancer, in a user / subject. The system in this particular embodiment comprises the disclosed screening device 100, a software application 121 installed on the user's computing device 120 (see Figure 4), a remote processing unit 130, and a database (or memory) 140. It should be noted that in other embodiments, the system need not include all of the above-mentioned elements. Furthermore, in other embodiments, the system may comprise more than one memory / database, and the memory / database 140 may be included in other system elements, i.e., either the software application 121, the first processing unit, and / or the remote processing unit 130, instead of being remote.

[0138] As shown in FIG. 3, the screening device 100, the software application 121, the remote processing unit 130, e.g., a cloud-based server, and the memory or database 140 are operatively connected and can exchange data between them (e.g., output signals or sensor data, data of the subject, data stored in the memory or database, i.e., health or demographic data of the subject and / or a number of healthy and unhealthy individuals, or disease-related results). In some embodiments, the screening device 100 is accessed via the user's computing device 120 and is then connected to the remote processing unit 130, which can transmit and receive data from the memory / database 140. Alternatively or complementary, data can further flow from the software application 121 to the remote processing unit 130 and from the remote processing unit 130 to the software application 121. Similarly, data can be evaluated and information can be exchanged directly from the user's computing device 120 connected to the remote processing unit 130. Additionally, the remote processing unit 130 may be connected to a memory / database 140, such that data flows from the remote processing unit 130 to the memory / database 140 and vice versa.

[0139] Thus, classification can be performed by the AI-based classification software 150 in many places in the proposed solution, such as the screening device 100, the software application 121, and / or the remote processing unit 130 (see dashed box in FIG. 4). The user / subject can view the results on the screening device 100 itself, for example via an output display, or on the user's computing device 120 using the software application 121. The user's computing device 120 can be any device, such as a smartphone, laptop computer, PDA, tablet computer, wearable device, or any combination of the aforementioned, among others.

[0140] The memory / database 140 may also store health and / or demographic data of the subject and multiple healthy and / or unhealthy individuals. For example, the health and / or demographic data of the user / multiple healthy and / or unhealthy individuals may include age, gender, sex, ethnicity, place of birth / residence, drug consumption (e.g., tobacco habits, alcohol habits, cannabis habits, other toxic habits), dietary habits, user or personal medical history, subject's family medical history, medication intake, family breast cancer history (e.g., mother, sister, daughter, grandmother, aunt), personal cancer history (e.g., breast cancer, lung cancer, ovarian cancer, colorectal cancer, melanoma, bladder cancer, non-Hodgkin's lymphoma, kidney cancer), weight, height, diabetes history, thyroid may include breast cancer, breast milk, polycystic ovary syndrome, osteoporosis, other endocrine disorders, timing of menarche, menstruation, menopause, number of children born by date of delivery, number of premature births, number of miscarriages, number of children born, reproductive status (e.g., pregnancy, breastfeeding, postpartum, fertility medications), history of breast disease (e.g., breast pain during menstrual cycle, hormonal changes, cysts, fibroadenomas, general lumps, breast discharge, sore, cracked and itchy nipples, inverted nipples), age at menarche, age at first birth, age at onset of menopause, number of mammograms and x-rays taken, etc.

[0141] In particular, to perform such classification, the present invention provides the cited artificial intelligence based classification software 150, which consists of processing and classification algorithms that utilize machine learning (ML) and / or artificial intelligence (AI) techniques. The classification is based on the nature of the human sample. Thus, the AI ​​based classification software 150 can determine the outcome regarding the disease by processing and classifying the generated output signal. In some other embodiments, the AI ​​based classification software 150 can further determine the outcome regarding the disease by considering a plurality of healthy and / or unhealthy individuals, as well as the health and / or demographic data of the user / subject.

[0142] In some embodiments, the disease result consists of an indication of the presence or absence of disease. For example, the result can only be a "0" or "1" value or a word (e.g., "healthy" or "unhealthy") that asserts whether the screening result is good or bad. In other embodiments, the result can further indicate the extent of disease (e.g., Level II), the type (or subtype) of disease (e.g., breast cancer for type of disease, or type of breast cancer for subtype of disease), a recommendation for treatment, or even the size of the tumor.

[0143] According to a particular embodiment, the AI-based classification software 150 includes an artificial neural network (ANN) that receives as input the demographic data of the user / subject. This ANN is used as an encoder to extract information about the medical condition of the subject and encodes this information so that it is readable and useful to undergo the next data processing steps. Furthermore, the sensor data (i.e., the generated output signal) can be fed to a long short-term memory (LSTM) network. The input layer of the LSTM network consists of two types of neurons: 1) neurons that receive as input the encoded subject's data encoded by the ANN, and 2) neurons that receive as input the data coming from the chemically sensitive sensors 201-209. Thus, the LSTM consists of two separate sets of neurons (two subnets). These are fully connected within themselves and with each other, i.e., they are essentially dense networks.

[0144] In one embodiment, the AI-based classification software 150 is executed by the software application 121. Thus, in this case, data (e.g., output signal or sensor data) flows from the screening device 100 to the computing device 120. According to this particular embodiment, the results generated by the AI-based classification software 150 are displayed on the computing device 120 by the software application 121. To generate the results, the AI-based classification software 150 may take into account a number of healthy and / or unhealthy individuals, as well as the health and / or demographic data of the user / subject, as explained above.

[0145] In another embodiment, the AI-based classification software 150 is executed by the remote processing unit 130. Thus, data (e.g., output signals) flow from the screening device 100 to the remote processing unit 130. Furthermore, additional data (e.g., subject data) collected by the computing device 120 can flow from the latter to the remote processing unit 130. Similarly, data stored in the memory / database 140 can flow from the latter to the remote processing unit 130 and vice versa. Results generated by the AI-based classification software 150 can be displayed on the computing device 120 by the software application 121, by the remote processing unit 130 itself, or by the screening device 100.

[0146] In yet another embodiment, the AI-based classification software 150 is executed by a first processing unit located in the screening device 100. In this case, the results generated by the AI-based classification software 150 can be displayed directly by the screening device 100, by the computing device 120, or by the remote processing unit 130, depending on the system / method configuration. To generate said results, the AI-based classification software 150 can take into account a number of healthy and / or unhealthy individuals, as well as health and / or demographic data of the user / subject, as explained above.

[0147] FIG. 5 shows another embodiment of the system. In this case, the first processing unit and the communication unit are implemented on a hardware board 110 included in the screening device 100. The two aforementioned components receive data (i.e., generated output signals) acquired in the analysis chamber 107 from the sample. The data can be classified on the screening device 100 or can be sent to the software application 121 or the remote processing unit 130. Both the software application 121 and the remote processing unit 130 can assign AI-based classification software 150. The software application 121 needs to be downloaded before using the screening device 100, and the data acquired from the test sample can be transferred to the software application 121 as soon as the communication unit receives the data. The software application 121 can then classify the data or send the data to the remote processing unit 130 for classification.

[0148] Prior to utilizing the AI-based classification software 150, the system checks for the latest version of the classification software 150 in the version bucket 131. Prior to any classification, the system ensures that the latest version of the classification algorithm 150 is installed. The AI-based classification software 150 periodically receives real-time updates. Once the data obtained from a given test sample has been analyzed and classified, these data are stored in the database 140. Furthermore, if the AI-based classification software 150 determines a diagnosis, this may be displayed as described above.

[0149] The software application 121 may be developed using Python and may run on Linux using a terminal session. The software application 121 may act as a user guide through the screening process. In some embodiments, access to the software application 121 is via a user login through a given user interface (see Figures 9 and 10). The software application 121 may further provide an application welcome screen, a menu, a personal data screen, and a screening history screen that provide various screens that a user of the application will see. The software application 121 may be used to obtain health and / or demographic data of the user / subject.

[0150] Working Example Example 1: First device prototype breast cancer screening setup The goal was to find VOC patterns in both synthetic and human urine from control subjects and metastatic breast cancer patients in order to train a screening device prototype that would eventually be able to screen for breast cancer in real human urine samples.

[0151] 1.1 Method 1.1.1 Synthetic urine preparation and human urine collection Six single VOC solutions and 30 synthetic urine samples were prepared, and 90 human urine samples were collected.

[0152] First, for the first theoretical approach, six VOC samples containing control VOCs (CTR) (i.e., VOCs typically found in healthy human urine) or breast cancer-related VOCs (BC) were prepared in 11 mL of distilled water (Table 1).

[0153] [Table 1]

[0154] After the single VOC solution aliquot, 30 synthetic urine samples were prepared to further test the classification algorithm. With the aim of making the synthetic urine as close to reality as possible, 10 types of urine were simulated by employing different VOC concentration values ​​within the expected range (Table 2). 11 mL of urine solution contained VOC concentrations corresponding to those expected to be encountered in 100 mL of human urine.

[0155] [Table 2]

[0156] Once the single VOC solutions and synthetic urine samples were prepared, human urine samples were collected. To ensure the anonymity of participants in the study, each participant was randomly assigned a number to which their sample would be referenced during analysis.

[0157] Human sampling was performed on 90 adult women. Patients were excluded if they had a history of any cancer type other than breast cancer, urinary tract infection, or other medical conditions that may affect urine composition. Alcohol consumption was suppressed within 12 hours prior to sample extraction. There were no exclusion criteria regarding patient ethnicity and tobacco, drug, or food consumption. Menstruation was not an exclusion criterion; however, its proximity to the sampling event was recorded to understand any future misclassification that may have occurred. Urine aliquots were transferred to 11 mL headspace vials and frozen until analysis. Human urine samples were stored at 0 °C for 24–48 hours.

[0158] Urine sampling of breast cancer patients was performed in 39 patients with advanced metastatic breast cancer at the Hospital Universitari Sant Joan, Reus (Spain). Their ages ranged from 29 to 75 years (mean age 54.9 ± 12.0 years). Patients were randomly selected by the hospital's breast oncology team. The protocol, patient consent and legal and ethical aspects were reviewed and approved by the Fundacio Institut d'Investigacio Sanitaria Pere Virgili Ethics Committee (CEIm).

[0159] A control human urine collection was performed on 51 subjects, uniformly sampled from Tarragona, Barcelona and Reus, within the age range of 18-78 years (mean age 29.7±15.8 years), who did not present or had no knowledge of any type of cancer. All subjects were excluded from diabetes, cardiovascular disease and other types of cancer and gave verbal informed consent before participation.

[0160] 1.1.2 Gas Chromatography-Mass Spectrometry (GC-MS) Sample Analysis The first consideration in the gas chromatography-mass spectrometry (GC-MS) data examination was to test whether the VOCs under study were actually detected by the analysis. GC-MS vacuum settings and machine calibrations were performed.

[0161] Single VOC solutions, synthetic urine and human samples were analyzed by GC-MS to better understand their composition. For the human samples, 51 control samples and 39 breast cancer samples were analyzed in triplicate. The GC-MS was programmed to perform 2052 analyses over a 20-minute period.

[0162] 1.1.3 Training and testing the classification algorithm In this step, previously obtained sample readings were checked. The objectives were as follows: 1) Examining the readings of single VOC solutions to determine whether the GC-MS and the instrument under study were able to adequately detect them. 2) Examining the synthetic urine readings to determine whether the GC-MS and the instrument under study were able to adequately distinguish between simulated diseased and simulated control urine, in other words, to assess whether urine composition is significant when multiple VOCs are simultaneously present. 3) To examine readings of human urine solutions to determine whether the GC-MS and the instrument under study can detect differences between classes.

[0163] Once sample preparation and GC-MS analysis were performed, pretreatment, processing and prediction algorithms were used to classify the samples. The input data were raw data files from the GC-MS containing odor profiles from VOC solutions, synthetic urine and human urine.

[0164] Signal pre-processing procedures were performed with the aim of turning the complex raw data into noise-free, readable and meaningful data. This was attempted by performing interpolation, chromatogram design, baseline removal, smoothing filtering, normalization and 3-fold sample averaging. This step was performed using an R-based code that takes the raw GM-MS files as input and processes them to output their total ion chromatograms (TICs).

[0165] For the processing step, a Python-based processing code was used, which takes the TICs as input and statistically processes them to output sample classifications. The algorithm used in this case was Principal Component Analysis (PCA) and a topological check was performed.

[0166] Next, the performance of the algorithm designed for prediction purposes was evaluated. Their results on single VOC preparations, synthetic samples, and human urine were evaluated separately. To evaluate whether GC-MS can detect the target VOCs, a VOC pattern comparator was obtained using R-based code and the k-nearest neighbor (kNN) algorithm.

[0167] 1.1.4 Equipment Prototype Design and Testing The next step consists of carrying out the sample procedure as previously described, but now, instead of using a GC-MS, the urine is analyzed using the in-vestigation device prototype. A first prototype of the device was designed and built, based on an Arduino board with four VOC chemically sensitive sensors (metal oxide sensors) to capture the "odor" of urine, called TIC. The metal oxide sensors used are: -TGS 2600, Figaro Inc. (sensitive to gaseous air pollutants) -TGS 2602, Figaro Inc. (sensitive to VOCs, NH3, methylbenzene, among others) -TGS 2610, Figaro Inc. ((particularly sensitive to propane and butane) -TGS 2620, Figaro Inc. (Sensitive to alcohol, organic vapors, and VOCs)

[0168] 90 human urine samples were sniffed by the device prototype and the previously designed software implemented on an Arduino microcontroller. Each sample was sniffed once by all four chemically sensitive sensors every second for 30 seconds. Thus, the output of the sample record is a 120-dimensional feature vector for every sample.

[0169] The input data to perform the classification were the odor profiles of synthetic urine and human samples sniffed by the device prototype. The classification algorithm was designed using PCA and kNN algorithms as templates.

[0170] Because the Arduino microprocessor does not have sufficient storage to host the PCA code, this code is designed online and then implemented in the Arduino, allowing it to run autonomously offline via another code implemented in the device, which contains the classification algorithm that allows the device to classify samples as control or breast cancer based on the VOC profile.

[0171] Once the algorithm was implemented on the Arduino microprocessor, 54 samples were randomly selected from the entire data set (n=90) and used to train the algorithm, i.e. to define the scores, i.e. the parameters, that define the PCA. The remaining 36 test samples were used to test the device prototype by classifying the samples. Finally, to validate the classification results obtained by the device prototype, the results were compared with the obtained GC-MS data (classification obtained by using PCA and kNN).

[0172] 1.2 Results 1.2.1 GC-MS screening and sample classification This section is dedicated to evaluating the classification success of human urine samples analyzed using GC-MS. As shown in Figure 6(A), the GC-MS data after dimensionality reduction by PCA visually shows the differences between control (cross) and metastatic breast cancer (triangle) urine samples. Figure 6(B) is an expanded view of Figure 6(A). Principal components (PCs) 1 and 2 explain most of the variance between the classes. PCs are new features that are constructed as linear combinations or mixtures of the initial features, i.e., the responses of the chemically sensitive sensors at each instant in time. Geometrically speaking, PCs represent the directions of the data that explain the maximum amount of variance, i.e., the lines that capture the most information of the data.

[0173] The score plot in Figure 6 shows that 81.7% of the samples in quadrant I (top left) are controls, 85.7% of the samples in quadrants II and III (right) are breast cancer, and 100% of the samples in quadrant IV (bottom left) are also breast cancer. Thus, we can say that dimensionality reduction performs a powerful optimization of the sample spatial variance, and the variance explained by the first two PCs can be largely caused by the class to which a given sample belongs.

[0174] This procedure aimed to determine whether GC-MS was sensitive enough to distinguish breast cancer samples from control samples, thus providing evidence that sufficient differences existed between the two classes of samples. Therefore, the first step of sample classification was performed using the GC-MS data.

[0175] After obtaining the score plot constructed by PCA for the analyzed human urine samples, the kNN algorithm was also used to obtain another classification. 46 training samples and 44 test samples were used. 11 out of 19 actual breast cancer samples were classified as breast cancer. 24 control samples were classified as such out of 25 control samples. As a result, this exact example works with a classification rate of 79.55%.

[0176] The next point to be considered for model validation concerns sample clustering and classification. Since the model operates on human samples that affect patient treatment and outcomes, not only classification rates but also false negative (sensitivity) and false positive (specificity) rates should be evaluated. The results are shown in Table 3.

[0177] [Table 3]

[0178] By relying on the above results, we can conclude that the kNN classification approach yields better class predictions than those of PCA, with kNN not only being superior in classification rate but also in sensitivity, which is one of the most important parameters when validating models in healthcare applications.

[0179] 1.2.2 Device prototype screening This section is dedicated to evaluating the success of classification of human urine samples analyzed with the device prototype. Each sensor reads each sample 30 times for each analysis, so the readings of each sample were presented in 4 columns (one per sensor) and 30 rows. Once the VOC patterns were obtained by analyzing the samples through the VOC chemically sensitive sensor, the obtained data were classified. 54 of the 90 samples were used to train the model, and the remaining 36 samples were used to test the model.

[0180] The score plot of the samples projection onto the plane PC1-PC3 is shown in more detail in Figure 7. By observing this projection, 33 samples cluster below the imaginary line, of which 29 (87.9%) are breast cancer samples. Above this threshold, there are 40 control samples, of which 85.1% are control samples. The breast cancer samples were found to be significantly different from the control samples.

[0181] Table 4 shows the validation results of the training model.

[0182] [Table 4]

[0183] From the test samples, the numbers of true positives (TP), true negatives (TN), false positives (FP) and false negatives (FN) were 12, 9, 11 and 4, respectively.

[0184] This initial model of an odor-based cancer screening device was further developed in Example 2.

[0185] Example 2: Breast cancer screening by using an improved device prototype 2.1 Method 2.1.1 Data collection and processing The examples were carried out using the data extracted from Example 1 (raw data from the four chemically sensitive sensors - not yet pre-processed). Thus, data from 51 urine samples from control subjects and 39 urine samples from breast cancer subjects were analyzed and classified using the improved classification algorithm.

[0186] Before classifying the data, the data was pre-processed to ensure that the classification would subsequently depend on the cancer features and not on irrelevant artifacts. First, the data was normalized.

[0187] 2.1.2 Software Design and Testing After sample preprocessing, the data were fed into a convolutional neural network (cNN). The net consisted of four convolution filters of sizes 32, 64, 128, and 128 and a linear activation. The output layer of the neural network presented two cells (two outputs) with a softmax activation function. Specifically, the model was trained on 65 urine samples, tested on 12 samples, and validated on 13 samples.

[0188] Other classification methods such as PCA, decision trees, kNN, and kNN with PCA were also implemented.

[0189] 2.1.3 Improved Equipment Prototype Design and Testing Compared to the previous prototype, this second device had nine VOC chemically sensitive sensors and a more powerful microprocessor. The statistical-based algorithms were also replaced by AI algorithms. Whereas the previous microprocessor was an Arduino UNO, this second prototype had an MKR Arduino WiFi, which included WiFi connectivity. In addition, the classification algorithms of this prototype were run in a cloud-based server, which allows for more computing power than the previous prototype, where the classification algorithms were run on the edge (on the Arduino board).

[0190] 2.2 Results As shown in Figure 8(A), the accuracy of the cNN model (y-axis) improves the more samples it is trained with, with each sample being used multiple times (x-axis). Since the accuracy follows a similar pattern in both the training ("*" line) and testing ("-" line) datasets, it can be concluded that the model is not overfitted. Similar logic (but this time applied to the loss function) can be applied to the loss function plot (Figure 8(B)). As the model is trained, the loss function decreases, and therefore classification becomes better. Numerical results are shown in Table 5.

[0191] [Table 5]

[0192] Other classification methods (PCA, decision trees, kNN, and kNN with PCA) were also performed, but none of these methods performed better than the A1 model (Table 6).

[0193] [Table 6]

[0194] Throughout the specification and claims, the terms "comprise" and "include" and variations thereof are not intended to exclude other technical features, additives, components, or steps. Additional objects, advantages, and features of the present invention will become apparent to those skilled in the art upon examination of the specification or may be learned by practice of the invention. Moreover, the present invention includes all possible combinations of specific and preferred embodiments described herein.

[0195] Although the present embodiments have been described with reference to specific exemplary embodiments, it will be apparent that various modifications and changes can be made to these embodiments without departing from the broader spirit and scope of the various embodiments.

Claims

1. 1. A system for screening for disease in a sample from a human subject, the system comprising: A portable medical screening device (100), comprising: A collection chamber (104) configured to collect at least one test sample from the human subject, the test sample comprising: urine; a set of chemically sensitive sensors (201-209) including a volatile organic compound sensitive sensor including a tin oxide sensor, the sensors detecting volatile organic compounds in the test sample and generating an output signal indicative of the presence or absence of volatile organic compounds in the test sample as a result of the detection; an analysis chamber (107) configured to allocate said set of chemically sensitive sensors (201-209); a first processing unit operatively connected to the set of chemically sensitive sensors (201-209) to receive the generated output signals; a communication unit operatively connected to said first processing unit and configured to establish communication with a remote processing unit and a software application installed on a computing device; - a software application (121) installed on a computing device (120) for collecting health and / or demographic data of said subjects and displaying results relating to said disease through a user interface; a remote processing unit (130) running an artificial intelligence based classification software (150) adapted to determine said outcome relating to said disease, The artificial intelligence based classification software (150) comprises a remote processing unit (130) selected from the group consisting of long short-term memory, convolutional neural networks, linear discriminant analysis, and K-nearest neighbors; a memory or database (140) containing health and / or demographic data of said subject and / or a plurality of healthy and / or unhealthy individuals; said artificial intelligence based classification software (150) merging said generated output signal with said health and / or demographic data of said subject and recognizing patterns in the information stored in said memory or database (140) to generate said result regarding said disease; The result relating to the disease includes an indication of the presence or absence of the disease, the disease being breast cancer.

2. 2. The system of claim 1, wherein the portable medical screening device (100) comprises a box (101) adapted to enclose the collection chamber (104), a grid (105) arranged above the collection chamber (104), the analysis chamber (107), the first processing unit, and the communication unit, the box (101) having a hollow portion for the introduction of a drawer (102).

3. The system of claim 1, wherein the health and / or demographic data of the subject is further stored in the memory or database (140), and the artificial intelligence based classification software (150) takes into account the stored health and / or demographic data of the subject to determine the outcome regarding the disease.

4. The system of claim 1, wherein the determined outcome regarding the disease is further stored in the memory or database (140), and the artificial intelligence based classification software (150) determines the outcome regarding the disease taking into account the stored outcome.

5. 2. The system of claim 1, wherein the health and / or demographic data of the subject and the plurality of healthy and unhealthy individuals is selected from the group consisting of age, sex, gender, ethnicity, place of birth / residence, weight, height, body mass index, medication taking, drug consumption, medical history, family medical history, cancer history, family cancer history, breast disease history, family breast cancer history, number of mammograms and x-rays performed, diabetes history, thyroid disorder, polycystic ovarian syndrome, osteoporosis, other endocrine disorders, menstrual, menarche and menopausal characteristics, reproductive status, number of children born to term, number of premature births, number of miscarriages, number of living children, number of children born, age at first birth, type of diet, eating habits, and combinations thereof.

6. The system of claim 1 , wherein the remote processing unit (130) is a cloud-based server.

7. The system according to any one of claims 1 to 6, wherein the software application (121) is located in a smartphone or a web browser.

8. 1. A method for screening for disease in a sample from a human subject, the method comprising: collecting at least one test sample from said human subject by a collection chamber (104) of a portable medical screening device (100), said test sample comprising urine; detecting volatile organic compounds in the test sample by a set of chemically sensitive sensors (201-209) including a volatile organic compound sensitive sensor including a tin oxide sensor assigned to an analysis chamber (107) of the portable medical screening device (100), and generating an output signal indicative of the presence or absence of volatile organic compounds in the test sample as a result of said detection; receiving, by a first processing unit of the portable medical screening device (100), the generated output signal; establishing communication with a remote processing unit (130) and a software application (121) installed on a computing device (120) by a communication unit operatively connected to the first processing unit; collecting health and / or demographic data of said subject via a user interface by a software application (121) installed on a computing device (120); determining an outcome regarding said disease by merging said generated output signal with said health and / or demographic data of said subject by an artificial intelligence based classification software (150) executed by a remote processing unit (130) and recognizing patterns in information stored in a memory or database (140) comprising health and / or demographic data of a plurality of healthy and unhealthy individuals, said artificial intelligence based classification software (150) is selected from the group consisting of long short-term memory, convolutional neural networks, linear discriminant analysis, and K-nearest neighbors; Displaying the results relating to the disease by the user interface of the software application (121) installed on a computing device (120), the result regarding the disease includes an indication of the presence or absence of the disease, the disease being breast cancer; and A method comprising:

9. 1. A portable medical device for screening for disease in a sample from a human subject, comprising: - a collection chamber (104) configured to collect at least one test sample of said human subject, said test sample comprising urine; a grid (105) placed above said collection chamber (104); - a drawer (102) for supporting, opening and closing said collection chamber (104); a set of chemically sensitive sensors (201-209) including a volatile organic compound sensitive sensor including a tin oxide sensor, said sensors detecting volatile organic compounds in said test sample and generating an output signal indicative of the presence or absence of volatile organic compounds in said test sample as a result of said detection; an analysis chamber (107) configured to allocate said set of chemically sensitive sensors (201 to 209); a first processing unit operatively connected to said set of chemically sensitive sensors (201-209) to receive said generated output signals; a communication unit operatively connected to said first processing unit and configured to establish communication with a remote processing unit (130) and with a software application (121) installed on a computing device (120), said remote processing unit (130) configured to execute an artificial intelligence based classification software (150) to determine an outcome regarding said disease, said software application configured to collect health and / or demographic data of said subject and to display said outcome regarding said disease via a user interface, said artificial intelligence based classification software (150) being configured to execute said artificial intelligence based classification software (150) to determine an outcome regarding said disease, said software application being configured to collect health and / or demographic data of said subject and to display said outcome regarding said disease via a user interface, a software (150) selected from the group consisting of long short-term memory, convolutional neural network, linear discriminant analysis, and K-nearest neighbors, configured to merge the generated output signal with the health and / or demographic data of the subject and recognize patterns in information stored in a memory or database (140) to generate the result regarding the disease, the memory or database (140) comprising health and / or demographic data of the subject and / or a plurality of healthy and / or unhealthy individuals, the result regarding the disease comprising an indication of the presence or absence of the disease, the disease being breast cancer; The portable medical screening device (100) is comprised of a box (101) adapted to enclose the collection chamber (104), the grid (105), the analysis chamber (107), the first processing unit, and the communication unit, the box (101) having a hollow portion for the introduction of the drawer (102).

Images