Portable electronic nose system for fast gynecological-conditions diagnosis in consulting room: A case study

Abstract

In this work, the analytical approach to fast diagnostics of gynecological conditions based on the screening of volatile compounds in cervical mucus using a portable electronic nose system was developed for the first time. The highly sensitive electronic nose system based on seven mass-sensitive piezo-sensors with nanostructured modifiers of electrodes was utilized for the detection of highly volatile biomolecules (amines, ketones, alcohols, aldehydes, carbonic acids) in cervical mucus samples directly in a consulting room. We were solving the task to develop a model for rapid identification of patients with gynecologic inflammation diseases according to results from gaseous sensors or a single sensor scanning of a mixture composition of volatile molecules released by mucus. Whereupon we propose the following group ranking as "conditionally healthy", "remission", and "inflammation/infection". Preliminary, by individual substances we estimated selectivity and kinetic features of sorption-desorption processes on the selected array of stabilized gaseous sensors. According to the groups of compounds the detection limit by an array of sensors is 20 ррm for ketones by acetone, 10 ppm for alcohols by ethanol, for arenas 2 ppm by toluene, 20 ppb by ammonia and for amines 5 ppb by diethylamine. For the correct interpretation of the "electronic nose" results, we used the doctor’s diagnosis, their description of the condition and the severity of the lesion, which was defined by the results of microbiological research in the laboratory. At the first stage, it was established that to ensure the reliability of solution-making during sample ranking into groups, time from the moment of biomaterial sampling to the measurement should not exceed 3–5 min. At the appointment in the doctor’s office and as part of a preventative medical check-up, there were analyzed 83 bioassays according to the developed method. To elaborate a model of gynecological status assessment, there were chosen 30 conditionally healthy patients (scheduled patient care) and the ones with deviations from the conditional norm, suggesting antibacterial treatment with various inflammation histories (active form, remission, and exacerbation). To make a decision, we utilized the full amount of data, which is achieved while measuring volatile molecules above mucus during sorption (80 sec) and spontaneous desorption (120 sec). The methods and results of experimental processing of patients’ data obtained directly in a consulting room are described below. The most informative sensors were identified applying an unsupervised model for processing the output curves in bio-sample vapours. The reduction of the number of sensors in the array to one was justified. There was proved high accuracy of mucus sample ranking by presence/ absence of gynecologic disease within mere 60 s. Sensors’ data processing was carried out using machine-learning methods without a coach but with the pseudo markup. We proposed a self-learning model for bioassay separation into three classes based on the signals from the most sensitive sensor with a classification error of no more than 10%. We conducted distribution into diagnostic groups and verification of the correct classification of new samples according to the results of microbiological analysis for microflora and key cells, according to the culture test method for infectious agents of sexually transmitted diseases (Gardnerella, Candida. alb, ureaplasmosis Ureap. Parvum, Trichomonas vaginalis). For accurate confirmation of pathogens’ presence, the polymerase chain reaction method was used. The model allows us to predict processes of remission and the presence of previous illnesses in remission, not necessarily infectious (e.g., adhesion formation). Active inflammation processes of any nature are also predicted well. Characteristics of the model for prediction of the presence of sexually transmitted infections by the signals of one sensor are sensitivity – 64%, and specificity – 86%. The capacity of an analysis based on the developed method and the model constitutes 30–35 samples per shift. The minimal lifetime of the studied sensors with bio-hydroxyapatite phase is more than 12 months. In such a case, there was neither sensor calibration nor their training on test substances required. A comparative study was conducted to validate the accuracy of using the electronic nose system versus classic cultural and polymerase chain reaction methods.