3D Reconstruction from biplanar x-ray
Team: Bishesh Khanal and Mahesh Shakya
Motivation: CT scans are often taken to visualize internal body organs, diagnose pathologies and for surgical planning. These devices provide better visualization than X-ray scans, but are also expensive, have high radiation dosage and acquisition time. But these devices may not always be available, especially in secondary and primary health care centers. For example, according to Nepal Health Facility Survey 2015, 96.6% of Zonal and above Hospitals had access to X-ray machines whereas only 41.3% had access to CT Scanners. This contrast is even more drastic for District-level Hospitals where none of them had access to CT Scanners but 85.5% had access to X-ray devices.
In that regard, we aim to develop algorithms for 3D reconstruction of select anatomical structures from biplanar x-ray images. This reconstruction is useful in various niche applications, providing better quantification of biomarkers and as a visual assistance in surgery rather than as a complete alternative to CT.
Research Direction: Various algorithms are available in the literature including classical Statistical Shape Model(SSM)-based to recent Deep learning-based ones. These algorithms do reasonably well in reconstructing normal/healthy anatomy, but have room for improvement in reconstructing pathological cases. We aim to improve this state-of-the-art focusing especially on reconstruction of bone structures with pathological conditions by exploring generative modelling based deep learning approach.