A granulometric constraint auto-encoder driven method for lung disease detection
By using a particle-sphere constrained autoencoder-driven method, the problems of insufficient feature extraction and insufficient ability to jointly diagnose multiple diseases in existing lung disease diagnosis are solved, enabling accurate detection and personalized treatment of lung diseases.
CN122245819APending Publication Date: 2026-06-19NANTONG UNIV
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANTONG UNIV
- Filing Date
- 2026-03-24
- Publication Date
- 2026-06-19
Smart Images
Figure CN122245819A_ABST
Abstract
This invention provides a lung disease detection method driven by a particle-sphere constrained autoencoder, comprising: step S10, processing a multi-label dataset of lung images; step S20, constructing a SoftSelector selection autoencoder; step S30, constructing particle-sphere constraints based on particle-sphere calculation; step S40, designing a fine-grained loss function for co-training; and step S50, iterative optimization. This invention effectively improves the feature selection quality of multi-label data containing missing labels. Through the co-optimization of the particle-sphere constrained autoencoder and the SoftSelector selection autoencoder, not only is the high discriminative power of the feature subset ensured, but the model's resistance to noise from missing labels is also enhanced. This method significantly reduces the probability of feature misselection for ambiguous samples and greatly improves the performance of downstream classification tasks, providing an efficient solution for the intelligent analysis of high-dimensional incomplete data.
Need to check novelty before this filing date? Find Prior Art