Aim of the Event

Integrating artificial intelligence methods with biomedical signals and medical images offers significant opportunities for the early identification of disease-related patterns, the development of data-driven decision support systems and the creation of personalised health technologies. Current studies show that AI-supported imaging systems can be used for image enhancement, classification, segmentation, anomaly detection and decision support; however, transferring these systems into clinical use requires explainability, standardised validation and integration with human expertise (Oulmalme et al., 2025; Wadie et al., 2026).

Developing a reliable smart diagnostic system does not depend on algorithmic knowledge alone. It also requires understanding the characteristics of biological systems, selecting a sensor appropriate to the physiological parameter to be measured, acquiring the data correctly, preserving signal quality, extracting meaningful features and evaluating model results against scientific criteria. Biomedical sensors combined with machine learning are reported to be increasingly important for monitoring physiological measurements and deriving meaningful indicators from health data (Gu et al., 2026; Zhao et al., 2026).

The core aim of this event is to provide participants with interdisciplinary, integrated and application-oriented knowledge and skills covering the whole smart-diagnostics development process — from measuring biological and physiological processes to analysing the resulting signals and images with artificial intelligence methods.

The first day covers the core components of smart diagnostic systems, the data flow from biomedical measurement to the AI model, the biophysical properties of biological systems, biomedical sensors and the lossless transmission of health data. The interaction of electromagnetic fields with biological tissue, RF-based imaging and the health applications of resonator-structured microwave sensors are also examined. At the end of the day, participants design a sample biomedical measurement system for a selected physiological parameter, covering sensor, sampling, signal conditioning, data acquisition and data transmission components.

The second day covers the analysis of physiological time series, feature extraction in the time and frequency domains, deep learning model development on biomedical data and the computer-aided analysis of dental pathology images. Recent systematic reviews show that deep learning and generative AI approaches carry significant potential for noise reduction, resolution enhancement, image quality improvement and the preservation of diagnostic information in medical images (Oulmalme et al., 2025), while reliable applications still require model validation, explainability and evaluation with real-world data (Wadie et al., 2026). The programme also teaches the working principle of fNIRS technology, neurovascular coupling, the measurement of HbO and HbR concentration changes, the removal of motion artefacts and physiological noise, feature extraction and classification with machine learning. Systematic reviews and meta-analyses indicate that using fNIRS data together with machine learning is promising for objective classification, but requires standard data collection, analysis, external validation and model explainability (De Giacomo et al., 2026; Li et al., 2026). The event further includes applications on vibration-based biomedical measurement and PEMF systems. The vibration session covers controlled excitation, data acquisition with an accelerometer, noise removal, the Fast Fourier Transform and the extraction of dominant frequency, amplitude, resonance and damping characteristics. The PEMF sessions examine a system consisting of a microcontroller, driver circuit, power supply and coil; pulse parameters are defined and the intensity and spatial distribution of the generated magnetic field are measured.

By the end of the training, participants are expected to be able to select a biomedical sensor and measurement method appropriate to a health problem; identify the sensor, sampling, signal conditioning and data acquisition components of a measurement system; prepare physiological signals and medical images for analysis; extract features in the time and frequency domains; explain the biomedical use principles of fNIRS, vibration, RF, microwave and PEMF systems; select an AI method suited to biomedical data; and design a sample smart diagnostic system that integrates measurement, data processing and artificial intelligence components.

Scope of the Event

The event is an interdisciplinary scientific training programme of two days and 20 course hours in total, covering biomedical measurement, sensor technologies, physiological signal and medical image analysis, and AI-based smart diagnostic applications. It is designed for undergraduate and graduate students, early-career researchers, academics and R&D staff in engineering, computer science, artificial intelligence and health sciences.

The first day addresses the core components of smart diagnostic systems, the biophysical properties of biological systems, biomedical sensors, data acquisition systems and the development of AI models suited to health problems. Measuring, digitising and processing biomedical data and preparing it for an AI model are treated as a single integrated data flow, on the premise that data quality in smart diagnostic systems is directly related to sensor selection, measurement conditions and data processing. The day also examines bioelectromagnetic measurement, RF imaging and the use of resonator-structured microwave sensors in health applications. Recent work shows that microwave sensors can detect changes in the electrical properties of biological samples without contact and with small sample volumes (Piekarz et al., 2024). Participants compare different physical sensing methods in terms of working principles, measurement reliability and biomedical applicability.

The second day covers the analysis of physiological time series, feature extraction in the time and frequency domains, deep learning and the computer-aided evaluation of medical images. Properties of biomedical time series such as differing sampling structures, limited labelled data, inter-individual variability and class imbalance require dedicated methods in the AI models built for them (Li et al., 2024). Recent research on physiological signals likewise shows the importance of preprocessing, feature extraction, model selection and validation for classification performance (Neifar et al., 2025; Jin et al., 2025). The programme covers measuring the haemodynamic response of the brain with fNIRS, obtaining HbO and HbR signals, removing motion artefacts and physiological noise, and evaluating the extracted features with machine learning. Studies combining fNIRS and machine learning show this to be an important research area for classifying clinical groups, while standardised measurement and analysis protocols are still needed (Eken et al., 2024). The vibration-based biomedical measurement session addresses the evaluation of the mechanical properties of biological tissue through controlled vibration and sensor data; participants examine the extraction of time- and frequency-domain features from sample vibration data and their classification with machine learning. The PEMF sessions focus on the electronic components of the system, signal parameters, field measurement, technical feasibility, repeatability and safety. Current studies show that PEMF effects depend on parameters such as frequency, field strength, waveform and application duration, so application protocols must be defined in a measurable and repeatable manner (Zahumenska et al., 2024; Picelli et al., 2024). PEMF is not applied as a treatment method during the event; it is examined solely within the scope of engineering system design and experimental field measurement.

The event is delivered through theoretical lectures, scientific discussion, case study analysis, data analysis, problem solving and system design. No clinical diagnosis or treatment is applied to humans or patients; practical work is limited to phantom models, anonymised sample data and engineering set-ups.