RGB to Hyperspectral: Improving Surgical Imaging
Author Information
Author(s): Tobias Czempiel, Alfie Roddan, Maria Leiloglou, Zepeng Hu, Kevin O'Neill, Giulio Anichini, Danail Stoyanov, Daniel Elson
Primary Institution: Imperial College London
Hypothesis
Can RGB data be effectively used to reconstruct hyperspectral signatures for enhanced surgical imaging?
Conclusion
The study demonstrates that transformer models significantly improve the accuracy of hyperspectral reconstruction from RGB data, which can enhance surgical decision-making.
Supporting Evidence
- Transformer models outperformed CNNs in reconstructing hyperspectral images from RGB data.
- Models like MST++ showed the best performance across various metrics.
- Reconstruction accuracy was higher for visible wavelengths compared to extended ranges.
Takeaway
This study shows that we can take regular color images and turn them into special images that show more details, helping doctors see better during surgery.
Methodology
The study evaluated various deep learning architectures, including CNNs and transformers, using two surgical datasets to assess the quality of spectral reconstruction.
Limitations
The study faced challenges in reconstructing extended spectral ranges and relied on the availability of specific datasets.
Participant Demographics
The study utilized datasets from porcine surgery and neurosurgery, involving 20 pigs and various surgical interventions.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website