Learn about pulmonary fibrosis, how AI is transforming lung disease detection, and the science behind our diagnostic engine.
Pulmonary fibrosis is a chronic, progressive lung disease where scarring (fibrosis) thickens the tissue around and between the air sacs (alveoli) in the lungs. This makes it harder for oxygen to pass into the bloodstream, leading to shortness of breath and reduced lung function over time.
The disease typically worsens over months to years, with a decline in lung capacity measured by Forced Vital Capacity (FVC). The rate of decline varies widely between patients, making accurate prognosis difficult. Early detection and monitoring are critical for guiding treatment and improving quality of life.
High-resolution computed tomography (HRCT) scans of the chest are essential for diagnosing pulmonary fibrosis. They reveal characteristic patterns such as honeycombing, ground-glass opacities, and traction bronchiectasis. These patterns help clinicians determine the type and severity of fibrosis.
PulmoScan leverages a 3D-ResNet + LSTM deep learning pipeline trained on thousands of CT volumes. The 3D-ResNet captures spatial features across slices, while the LSTM models temporal and sequential relationships. This architecture achieves state-of-the-art performance in predicting FVC decline and classifying fibrosis severity.
Early identification of pulmonary fibrosis allows for timely intervention with antifibrotic therapies that can slow progression. Without treatment, the disease can advance to respiratory failure. AI-assisted screening can flag subtle changes in CT scans that may be missed in routine readings.
PulmoScan was developed using the OSIC Pulmonary Fibrosis Progression dataset (Kaggle 2020) and the Chest CT-Scan Images Dataset by Mohamed Hany. These datasets provide diverse training data covering various demographics, scanner types, and disease stages — enabling robust and generalizable AI predictions.