A FiLM-Enhanced DenseNet Approach for Multi-Label Chest X-ray Analysis Using Patient Metadata
Year: 2025
Type: preprint
Abstract: Accurate diagnosis of thoracic diseases from chest X-rays is crucial for effective patient care, but variability between radiologists and limited access to experts can make this challenging. In this study, we present a deep learning framework that combines chest X-ray images with patient information such as age, gender, and view position. The framework uses Feature-wise Linear Modulation layers integrated into a DenseNet-121 backbone to incorporate this metadata directly into the image analysis process. This work builds on the Doctor Ji project and focuses on improving both diagnostic accuracy and interpretability in practical clinical settings. We evaluated the model on a curated subset of 15,000 images from the NIH ChestX-ray14 dataset. The FiLM-augmented DenseNet-121 achieved a Micro AUC of 0.882 and a Macro AUC of 0.741, outperforming strong baselines including ResNet-50, ConvNeXt-Tiny, and ViT Base. Grad-CAM visualizations showed that the model focuses on clinically relevant regions, providing insights into the decision-making process. This study demonstrates a reproducible approach for integrating patient context into automated chest X-ray diagnosis and highlights the potential of combining image and metadata information to enhance clinical AI tools.
Author: Aditya Gupta
Institution: Bennett University
Language: English
Topic: COVID-19 diagnosis using AI
Subfield: Radiology, Nuclear Medicine and Imaging
Field: Medicine
Domain: Health Sciences
Open Access status: gold
OpenAlex ID: W4414787967
Cites:
Cited by:
Related to: