A model training method and device for lung disease diagnosis
By collaboratively designing customized data preprocessing, hybrid loss functions, and refined training strategies, the shortcomings of training data processing and model training in lung disease diagnosis are addressed, resulting in a high-precision and stable lung disease diagnostic model that meets the needs of real-time bedside diagnosis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- XIAN CHEST HOSPITAL (XIAN INSTITUTE OF RESPIRATORY DISEASES)
- Filing Date
- 2026-04-29
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies for diagnosing lung diseases suffer from problems such as a lack of targeted training data processing, independent design of each stage of model training, and insufficient connection between training methods and clinical needs. These issues result in insufficient segmentation accuracy and stability, failing to meet the needs of real-time bedside diagnosis.
A collaborative design of customized data preprocessing, hybrid loss function configuration, and refined training strategies is adopted, including balanced dataset partitioning, ConvXt-UNet architecture construction, hybrid loss functions of FocalLoss and TverskyLoss, adaptive gradient optimizer, and step-wise learning rate decay. Combined with dual-metric monitoring, the model can be trained accurately.
The model improved segmentation accuracy and stability, meeting the needs of real-time clinical diagnosis, enhancing consistency between the model and annotations by senior physicians, reducing misdiagnosis and missed diagnosis, and achieving standardized diagnosis of lung diseases.
Smart Images
Figure CN122416183A_ABST