A convolutional neural network approach for automated quantification of emphysema and fibrosis in idiopathic interstitial pneumonia (IIP)

Congress:

ECR 2022

Poster Number:

C-14712

Type:

Scientific Exhibit

Keywords:

Artificial Intelligence, CT-High Resolution, Computer Applications-Detection, diagnosis, Tissue characterisation

Authors:

C. H. Leow, V. Corona, P. Yousefi, M. Purtorab, K. Tait, S. Mohammadi, B. Irving, G. Brüggenwerth

DOI:

10.26044/ecr2022/C-14712

DOI-Link:

https://dx.doi.org/10.26044/ecr2022/C-14712

Conclusion

Conclusion

Manual delineation of emphysema and fibrosis is challenging. Our deep learning approach shows promising results in the automatic quantification of pulmonary fibrosis and emphysema. Automatic delineation of fibrotic and emphysematous regions for the entire volume in less than a minute represents a key advantage of our approach. Such techniques can be used to better understand the extent and progression of disease through HRCT, to help identify patient population in clinical trials and to better inform treatment decisions, thus improving patient outcomes.

Future studies in larger prospective cohorts are needed to further confirm the clinical practicality.

Fig 1: Correlation between deep learning (DL) approach and conventional baseline...

Fig 2: Correlation plots for Analyst vs. DL.

Fig 3: Comparison of raw image (top left) to ground truth (top right), baseline...

Fig 4: 3D surface rendering of the DL prediction from Figure 3 (bottom right) with...

Fig 5: Comparison of raw image (top left) to ground truth (top right), baseline...

Fig 6: Comparison of raw image (top left) to ground truth (top right), baseline...