2023 · 12 citations · 26 references
ComputingComputer ScienceHealth InformaticsDiagnostic ReasoningMachine LearningPulmonologyChronic Obstructive Pulmonary DiseaseDeep LearningExplainable Machine LearningObstructive Lung DiseasePulmonary ImagingRepresentation LearningData AugmentationDeep Learning DiagnosticsExplainabilityExplainable Chest X-ray InterpretationModel Performance AnalysisPatient Outcomes MeasurementPredictive Model PerformanceVisual Explanations
Chronic Obstructive Pulmonary Disease (COPD) is a prevalent pulmonary condition marked by enduring respiratory symptoms and airflow limitations. Prompt diagnosis is vital for efficacious disease management and enhanced patient outcomes. This study introduces COPDNet, an explainable deep learning model founded on the ResNet50 architecture, tailored specifically for diagnosing COPD using Chest radiographic (CXR) images. COPDNet employs preprocessing techniques such as normalization, resizing, and data augmentation to boost model performance and generalization. To foster interpretability and clinician trust, explainability methods like Layer-wise Relevance Propagation (LRP) or Grad-CAM are integrated, offering visual explanations for model predictions. In addition, transfer learning is utilized by initializing the model with pre-trained weights from extensive datasets, minimizing training time and refining performance. With a recall of 98.9% on the pneumothorax CXR image, the efficacy of COPDNet in diagnosing COPD was validated, potentially leading to improved patient outcomes and supporting clinicians in making well-informed decisions.
26
ImageNet Large Scale Visual Recognition Challenge
Olga Russakovsky, Jia Deng, Hao Su et al. · International Journal of Computer Vision · 2015 · 40.2K citations
Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization
Ramprasaath R. Selvaraju, Michael Cogswell, Abhishek Das et al. · 2017 · 21.6K citations
Hoo-Chang Shin, Holger R. Roth, Mingchen Gao et al. · IEEE Transactions on Medical Imaging · 2016 · 5.8K citations · Full text