Research On Metabolic Syndrome Diagnosis Algorithm Based On Human Infrared Data

The common chronic diseases closely related to metabolic disorder,such as type 2 diabetes,coronary heart disease and stroke,are serious hazards to national health.They are the key issues of clinical research in recent years,and the metabolic syndrome,an important development node in its early stage,has also become the focus of research.Medical infrared thermal imaging technology is a green non-invasive structural image detection method,which can detect the temperature distribution of the patient’s body for medical diagnosis.It is one of the main diagnostic methods of metabolic syndrome.At present,the disease diagnosis research based on infrared thermal imaging technology at home and abroad is mainly limited to breast cancer,thyroid disease and other diseases with obvious local abnormal characteristics,and relies on a single symmetry analysis,ignoring other characteristics of human body temperature distribution,such as uniformity of temperature distribution and thermal sequence,which are also crucial for disease diagnosis.Therefore,this article proposes a metabolic syndrome auxiliary diagnosis model based on multi characteristic quantification of human infrared data,and combines it with a few-shot analysis algorithm based on distillation networks to enhance the model’s feature extraction ability to address issues such as unclear features and sparse samples in infrared image data.The main work of this thesis includes:1.A fusion multi feature quantification assisted diagnostic model is proposed for diseases with significant changes in local features in human infrared data research.Firstly,modeling and analyzing symmetry,uniformity,and sequence separately:conducting symmetry analysis based on regional high-order statistical parameter distribution similarity,quantifying uniformity based on the overall connectivity and patch measurement of infrared data,and conducting sequence analysis using labeled clustering algorithms.Then,for diseases such as metabolic syndrome that are not limited to a specific area of the body but exhibit systemic changes,multiple feature fusion is proposed to further improve diagnostic accuracy.2.In response to the problem of small sample size and insufficient feature extraction,a few-shot learning based on distillation network is proposed.This algorithm gradually analyzes data through multiple modules to enhance the model’s feature extraction ability.In the model,the inception block is embedded to improve the multi-scale adaptability of the network to few-shot data.Through distillation training,the small network can learn parameters from the large network to ensure the performance of the model while effectively avoiding overfitting.Finally,the similarity comparison of the model output feature vectors is used to classify to achieve accurate diagnosis of patients with metabolic syndrome.