Intelligent detection model training method and system for respiratory system disease image
By constructing a teacher identification model and a lightweight student identification model, and utilizing a knowledge distillation framework, efficient and rapid lung CT image detection was achieved on primary healthcare equipment. This resolved the contradiction between detection accuracy and deployment feasibility, and improved the auxiliary detection efficiency and diagnostic interpretability of early lesions.
CN122244583APending Publication Date: 2026-06-19张丽丽
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 张丽丽
- Filing Date
- 2026-02-10
- Publication Date
- 2026-06-19
Smart Images
Figure CN122244583A_ABST
Abstract
This application provides a method and system for training an intelligent detection model for respiratory disease images. It acquires labeled lung CT image sequences to construct a training dataset, builds a teacher recognition model, and constructs a lightweight student recognition model. During the training phase, a knowledge distillation framework is used to drive the student recognition model to learn the true labels of the training dataset. Simultaneously, it mimics the softened probability distribution and intermediate layer feature responses output by the teacher recognition model, which are rich in spatial and channel attention information. The trained student recognition model is then used to infer the lung CT images of patients with respiratory diseases to be diagnosed, obtaining the detection results of the lung CT images of these patients. Using the scheme of this application, intelligent detection of lung CT images of respiratory patients can be performed based on a dual attention mechanism and knowledge distillation.
Need to check novelty before this filing date? Find Prior Art