On AI-Assisted Pneumoconiosis Detection from Chest X-rays

Yasmeena Akhter, Rishabh Ranjan, Richa Singh, Mayank Vatsa, Santanu Chaudhury

Introduction

Pneumoconiosis, a group of irreversible occupational lung diseases including silicosis and "black lung," remains a significant public health crisis, particularly in the mining and construction sectors of developing nations like India. These conditions are often misdiagnosed as Tuberculosis (TB) due to their visual similarity on standard chest X-rays (CXRs). This diagnostic challenge is compounded by a severe shortage of expert radiologists in remote mining regions. While AI has advanced significantly in detecting common respiratory diseases, occupational lung diseases are frequently under-represented in standard datasets, leading to suboptimal performance by general-purpose medical AI models.

The Proposed AI-Assisted Pipeline

To address these gaps, researchers have developed a multi-stage "Digital Diagnostic Assistant" designed to provide explainable and actionable insights for healthcare workers. The pipeline moves beyond simple binary classification to offer a comprehensive diagnostic tool.

The framework consists of several key components:

  1. Preprocessing & Segmentation: Enhances image quality and isolates lung regions to reduce noise.
  2. Abnormality Localization: Utilizes Mask R-CNN to identify and bound specific pathological findings such as nodules or opacities. This "local labeling" ensures clinicians can verify the AI's focus areas.
  3. Disease Prediction: Employs CheXNet to categorize images into four distinct classes: Normal, TB, Silicosis, or Silicotuberculosis.
  4. Automated Reporting: A GPT-based Transformer architecture converts localized findings into natural language medical reports, bridging the expertise gap for non-specialist healthcare workers.
Overview of the AI-assisted diagnostic pipeline, from input image to automated report generation.

The Sirohi Dataset

A critical contribution of this work is the introduction of the Sirohi Dataset, specifically tailored to the Indian context. It comprises 4,077 CXR samples collected from primary healthcare centers in the Sirohi district of Rajasthan. The dataset was meticulously annotated by expert radiologists from Tele-Radiology Solutions Bangalore and AIIMS Jodhpur. It includes both global disease labels and localized annotations for findings like lung opacities and pleural effusion, providing a robust foundation for training specialized models.

Experimental Results and Insights

Baseline evaluations were conducted using three prominent deep learning architectures: ResNet18, DenseNet121, and InceptionV3. The results highlight the inherent difficulty in distinguishing between pure Silicosis and pure TB, a phenomenon referred to as the "TB Mimicry Trap."

Performance metrics of baseline deep learning models across different diagnostic classes.

As shown in the performance table, while models achieved high AUC scores for "Normal" cases (up to 0.91), performance dipped significantly for TB (as low as 0.71). This underscores the necessity of the specialized, explainable architecture proposed in this study. The high inter-class similarity between early-stage silicosis and TB remains the primary hurdle for automated systems.

Conclusion and Future Directions

While the proposed framework shows promise, several challenges remain. Deep learning models require even larger, more diverse datasets to achieve clinical-grade reliability. Furthermore, deploying AI in rural settings necessitates overcoming infrastructure constraints such as inconsistent power and internet connectivity.

Future work will focus on refining the model's ability to distinguish between various pneumoconiosis variants and integrating the tool into broader telemedicine frameworks. By combining automated localization with natural language reporting, this AI-assisted approach aims to empower local healthcare providers and improve outcomes for vulnerable worker populations.

Summary updated

Abstract

According to theWorld Health Organization, Pneumoconiosis affects millions of workers globally, with an estimated 260,000 deaths annually. The burden of Pneumoconiosis is particularly high in low-income countries, where occupational safety standards are often inadequate, and the prevalence of the disease is increasing rapidly. The reduced availability of expert medical care in rural areas, where these diseases are more prevalent, further adds to the delayed screening and unfavourable outcomes of the disease. This paper aims to highlight the urgent need for early screening and detection of Pneumoconiosis, given its significant impact on affected individuals, their families, and societies as a whole. With the help of low-cost machine learning models, early screening, detection, and prevention of Pneumoconiosis can help reduce healthcare costs, particularly in low-income countries. In this direction, this research focuses on designing AI solutions for detecting different kinds of Pneumoconiosis from chest X-ray data. This will contribute to the Sustainable Development Goal 3 of ensuring healthy lives and promoting well-being for all at all ages, and present the framework for data collection and algorithm for detecting Pneumoconiosis for early screening. The baseline results show that the existing algorithms are unable to address this challenge. Therefore, it is our assertion that this research will improve state-of-the-art algorithms of segmentation, semantic segmentation, and classification not only for this disease but in general medical image analysis literature.
BibTeX
@inproceedings{akhter2023ai,
  author = {Yasmeena Akhter and Rishabh Ranjan and Richa Singh and Mayank Vatsa and Santanu Chaudhury},
  title = {{On AI-Assisted Pneumoconiosis Detection from Chest X-rays}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2023},
  doi = {10.24963/ijcai.2023/705},
}
Figure