Image prediction of disease progression for osteoarthritis by style-based manifold extrapolation
Disease-modifying management aims to prevent deterioration and progression of the disease, and not just to relieve symptoms. We present a solution for the management by a methodology that allows the prediction of progression risk and morphology in individuals using a latent extrapolation approach. T...
Saved in:
| Main Authors: | , , , , , , , , , , , , , |
|---|---|
| Format: | Article (Journal) |
| Language: | English |
| Published: |
16 November 2022
|
| In: |
Nature machine intelligence
Year: 2022, Volume: 4, Issue: 11, Pages: 1029-1039 |
| ISSN: | 2522-5839 |
| DOI: | 10.1038/s42256-022-00560-x |
| Online Access: | Verlag, lizenzpflichtig, Volltext: https://doi.org/10.1038/s42256-022-00560-x Verlag, lizenzpflichtig, Volltext: https://www.nature.com/articles/s42256-022-00560-x 
|
| Author Notes: | Tianyu Han, Jakob Nikolas Kather, Federico Pedersoli, Markus Zimmermann, Sebastian Keil, Maximilian Schulze-Hagen, Marc Terwoelbeck, Peter Isfort, Christoph Haarburger, Fabian Kiessling, Christiane Kuhl, Volkmar Schulz, Sven Nebelung, Daniel Truhn |
| Summary: | Disease-modifying management aims to prevent deterioration and progression of the disease, and not just to relieve symptoms. We present a solution for the management by a methodology that allows the prediction of progression risk and morphology in individuals using a latent extrapolation approach. To this end, we combined a regularized generative adversarial network and a latent nearest neighbour algorithm for joint optimization to generate plausible images of future time points. We evaluated our method on osteoarthritis data from a multicenter longitudinal study (the Osteoarthritis Initiative). With presymptomatic baseline data, our model is generative and considerably outperforms the end-to-end learning model in discriminating the progressive cohort. Two experiments were performed with seven radiologists. When no synthetic follow-up radiographs were provided, our model performed better than all seven radiologists. In cases in which the synthetic follow-ups generated by our model were made available to the radiologist for diagnosis support, the specificity and sensitivity of all readers in discriminating progressors increased from 72.3% to 88.6% and from 42.1% to 51.6%, respectively. Our results open up a new possibility of using model-based morphology and risk prediction to make predictions about disease occurrence, as demonstrated by the example of osteoarthritis. |
|---|---|
| Item Description: | Gesehen am 18.01.2023 |
| Physical Description: | Online Resource |
| ISSN: | 2522-5839 |
| DOI: | 10.1038/s42256-022-00560-x |