Multimodal data driven machine learning approach for enhancing diagnostic accuracy and efficiency in pulmonary...

The management of pulmonary diseases in primary healthcare settings is often challenged by heterogeneous clinical evidence, incomplete patient information, and the need for reliable diagnostic support under limited resources. Conventional diagnostic approaches frequently rely on isolated information sources or static prediction procedures, which may restrict their ability to integrate diverse evidence and represent uncertainty in real clinical environments.

To address this problem, this study proposes a multimodal data driven machine learning framework, termed the Probabilistic Agent Driven Diagnostic Model (PADDM), for pulmonary related diagnostic support in primary healthcare scenarios. The framework integrates three coordinated components, namely the Manifold Constrained Feature Encoder, the Agent Based Decision Planner, and the Uncertainty Propagation Regularizer. These components are designed to support multimodal representation learning, finite step diagnostic belief refinement, and uncertainty aware prediction within a unified optimization framework. The proposed method was evaluated across heterogeneous pulmonary related data resources covering respiratory physiological signals, primary care symptom records, algorithm derived diagnostic evidence, and healthcare interaction logs.

Experimental results showed improved accuracy, precision, recall, and F1 score compared with the evaluated baseline methods under the same experimental settings. Additional analyses, including ablation experiments, confusion matrix evaluation, calibration assessment, inference latency, parameter count, and memory usage, further suggest that the proposed framework may improve diagnostic reliability and maintain a reasonable computational cost under the tested conditions. These findings indicate the potential of the proposed framework for supporting pulmonary diagnostic decision making in resource constrained primary healthcare settings. Further validation using larger, externally verified, prospectively collected, and clinically harmonized multimodal datasets is needed before broader clinical deployment can be established.