Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Imaging-Based Mortality Prediction in Patients with Systemic Sclerosis

Alec K. Peltekian, Karolina Senkow, Gorkem Durak, Kevin M. Grudzinski, Bradford C. Bemiss, Jane E. Dematte, Carrie Richardson, Nikolay S. Markov, Mary Carns, Kathleen Aren, Alexandra Soriano, Matthew Dapas, Harris Perlman, Aaron Gundersheimer, Kavitha C. Selvan, John Varga, Monique Hinchcliff, Krishnan Warrior, Catherine A. Gao, Richard G. Wunderink, GR Scott Budinger, Alok N. Choudhary, Anthony J. Esposito, Alexander V. Misharin, Ankit Agrawal, Ulas Bagci

TL;DR

This work tackles the challenge of predicting mortality in systemic sclerosis–associated interstitial lung disease using longitudinal chest CT data. It introduces a large-scale framework that compares radiomics and deep learning approaches, including a transformer-based model (Swin Transformer), on 2,125 scans with mortality labels at 1, 3, and 5 years. Deep learning models, especially Swin Transformer variants, outperform radiomics across timeframes, achieving peak AUROC values of up to $0.801$ for 3-year mortality, with DenseNet-121 excelling at 5-year mortality ($0.709$). The study demonstrates the feasibility and value of AI-driven, imaging-based risk stratification in SSc-ILD, offering a blueprint for multimodal and longitudinal prognostic tools in rare fibrotic diseases.

Abstract

Interstitial lung disease (ILD) is a leading cause of morbidity and mortality in systemic sclerosis (SSc). Chest computed tomography (CT) is the primary imaging modality for diagnosing and monitoring lung complications in SSc patients. However, its role in disease progression and mortality prediction has not yet been fully clarified. This study introduces a novel, large-scale longitudinal chest CT analysis framework that utilizes radiomics and deep learning to predict mortality associated with lung complications of SSc. We collected and analyzed 2,125 CT scans from SSc patients enrolled in the Northwestern Scleroderma Registry, conducting mortality analyses at one, three, and five years using advanced imaging analysis techniques. Death labels were assigned based on recorded deaths over the one-, three-, and five-year intervals, confirmed by expert physicians. In our dataset, 181, 326, and 428 of the 2,125 CT scans were from patients who died within one, three, and five years, respectively. Using ResNet-18, DenseNet-121, and Swin Transformer we use pre-trained models, and fine-tuned on 2,125 images of SSc patients. Models achieved an AUC of 0.769, 0.801, 0.709 for predicting mortality within one-, three-, and five-years, respectively. Our findings highlight the potential of both radiomics and deep learning computational methods to improve early detection and risk assessment of SSc-related interstitial lung disease, marking a significant advancement in the literature.

Imaging-Based Mortality Prediction in Patients with Systemic Sclerosis

TL;DR

This work tackles the challenge of predicting mortality in systemic sclerosis–associated interstitial lung disease using longitudinal chest CT data. It introduces a large-scale framework that compares radiomics and deep learning approaches, including a transformer-based model (Swin Transformer), on 2,125 scans with mortality labels at 1, 3, and 5 years. Deep learning models, especially Swin Transformer variants, outperform radiomics across timeframes, achieving peak AUROC values of up to for 3-year mortality, with DenseNet-121 excelling at 5-year mortality (). The study demonstrates the feasibility and value of AI-driven, imaging-based risk stratification in SSc-ILD, offering a blueprint for multimodal and longitudinal prognostic tools in rare fibrotic diseases.

Abstract

Interstitial lung disease (ILD) is a leading cause of morbidity and mortality in systemic sclerosis (SSc). Chest computed tomography (CT) is the primary imaging modality for diagnosing and monitoring lung complications in SSc patients. However, its role in disease progression and mortality prediction has not yet been fully clarified. This study introduces a novel, large-scale longitudinal chest CT analysis framework that utilizes radiomics and deep learning to predict mortality associated with lung complications of SSc. We collected and analyzed 2,125 CT scans from SSc patients enrolled in the Northwestern Scleroderma Registry, conducting mortality analyses at one, three, and five years using advanced imaging analysis techniques. Death labels were assigned based on recorded deaths over the one-, three-, and five-year intervals, confirmed by expert physicians. In our dataset, 181, 326, and 428 of the 2,125 CT scans were from patients who died within one, three, and five years, respectively. Using ResNet-18, DenseNet-121, and Swin Transformer we use pre-trained models, and fine-tuned on 2,125 images of SSc patients. Models achieved an AUC of 0.769, 0.801, 0.709 for predicting mortality within one-, three-, and five-years, respectively. Our findings highlight the potential of both radiomics and deep learning computational methods to improve early detection and risk assessment of SSc-related interstitial lung disease, marking a significant advancement in the literature.

Paper Structure

This paper contains 12 sections, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methods
  3. Image Data Collection
  4. Image Preprocessing and Standardization
  5. Radiomics Processing
  6. Deep Learning Segmentation of Lungs
  7. Mortality Prediction Modeling
  8. Results
  9. Imaging Data Overview
  10. Survival Trends and Mortality Risk
  11. Predictive Modeling Results
  12. Discussion and Conclusion

Figures (4)

  • Figure 1: Longitudinal CT scans illustrating ILD progression in three patients with SSc, with red boxes highlighting disease progression. P1 shows a patient who remained ILD-negative throughout the surveillance period. P2 shows a patient who developed ILD over time, with images taken at baseline and the 3rd, 10th, and 16th years, demonstrating disease onset and progressive fibrotic changes (note increasing reticular patterns in highlighted regions). P3 presents a patient with pre-existing ILD that rapidly worsened, with scans captured at baseline and the 1st, 2nd, and 3rd years, showing significant progression of fibrotic remodeling and honeycombing patterns within the marked areas.
  • Figure 2: Overview of the data preprocessing pipeline. (Data Preparation) All CT scans were resampled to a uniform 1mm slice thickness before feature extraction and segmented with LungMask R231 Model. (Radiomics) Radiomic features were extracted from the original/segmented lung pair, and used for training machine learning models using optuna hyperpameter optimization, selecting the best model with optimal parameters for each task. (Deep Learning) Segmented lung regions, multiplied with original CT images, are used as input for deep learning modeling.
  • Figure 3: CT scan utilization and ILD survival patterns.(A) Annual distribution of CT scans from the Northwestern Scleroderma Registry (2001–2023), illustrating an increasing trend in CT imaging utilization. (B) Distribution of survival times from the first ILD-confirmed CT scan to death, demonstrating significant variability in survival outcomes.
  • Figure 4: Mortality timing relative to CT scans.(A) Time from a patient’s first ILD-confirmed CT scan to death, emphasizing variability in long-term survival after initial ILD detection. (B) Time from the last recorded CT scan to death, showing that most patients undergo imaging within a specific window before mortality