The invention discloses a potential
nephropathy prediction electronic
nose system and method based on
machine learning, and belongs to the technical field of medical detection. According to the
system,
ammonia concentration and temperature and
humidity parameters in
exhaled air of a
human body are collected through a multi-mode sensor module, baseline calibration, flow velocity compensation and
noise filtering are carried out in combination with a data preprocessing module, and a standardized
characteristic matrix is generated. The
machine learning prediction model adopts a Bayesian
deep learning or integrated learning
algorithm, inputs an
ammonia concentration
peak value, an integral area and temperature and
humidity parameters, outputs a
nephropathy risk probability and stage prediction, and quantifies key influence factors through an SHAP value (for example, a high-risk alarm is triggered when an
ammonia concentration contribution degree is greater than or equal to 60%). The
system is equipped with a real-time GUI
interaction interface, dynamically displays a risk thermodynamic diagram and multi-dimensional data
overlay analysis, and supports clinicians to feed back and correct. And the data iteration module realizes
continuous optimization of the model through an encrypted
cloud storage and
incremental learning framework (Online SGD). A heterogeneous
sensor array (such as a Fe2Mo3O8 / MoO2 (at) MoS2
composite material) and a dynamic temperature and
humidity compensation technology are adopted, so that the detection stability is remarkably improved, and the clinical misdiagnosis rate is reduced. The system can be integrated in a portable device or a
hospital system, is suitable for
community screening, emergency
triage and remote medical scenes, and has the advantages of noninvasive, high-precision and cross-mechanism collaborative diagnosis.