Engineering Leader in Healthcare AI | Building Scalable, Data-Driven Solutions | Leading Cross-Functional Teams in Healthcare Innovation
Turning 2D scans into 3D reconstructions is key for many clinical applications, and shape models are an important part of the solution. As we highlighted in our recent #Kitware blog, statistical shape models (SSMs) can help address limitations such as sparse views, occlusions, and imaging variability by incorporating anatomical context into the reconstruction process. 🔗 https://lnkd.in/eDJvgZJc The potential clinical impact spans multiple specialties: 🦴 Orthopedics: 3D skeletal models from X-rays support fracture evaluation, deformity correction, and surgical planning at lower cost and radiation than CT. ❤️ Cardiovascular medicine: 3D models from echocardiography or angiography improve visualization for valve repair, stent placement, and aneurysm treatment. 🧠 Neurosurgery: X-ray-based reconstructions enable 3D views of tumors, spinal anatomy, and neural structures for safer, more precise planning. 😁 Dental & maxillofacial surgery: Transforming panoramic or cephalometric X-rays into 3D aids orthodontics and implant placement while reducing cone-beam CT scans. 🩺 GI & urology: Endoscopic video can be reconstructed into 3D models of the colon, bladder, or prostate, supporting minimally invasive interventions. ⛑️ Trauma & emergency medicine: Portable X-rays combined with reconstruction algorithms provide 3D insight when CT isn’t available, supporting urgent decisions. In our experience using SlicerSALT and ShapeWorks for these problems, we have seen how shape analysis can strengthen reconstruction pipelines and make them more clinically meaningful. We would love to hear your perspective. Where do you see shape models making the biggest impact in 2D-to-3D reconstruction for medical imaging? Let us know in the comments or reach out to us to continue the conversation Dženan Zukić Jared Vicory Beatriz Paniagua #MedicalImaging #ShapeAnalysis #HealthcareAI #DigitalHealth #OpenSource #SlicerSALT #ShapeWorks #3DReconstruction
SSMs are just linear generative models. This is 20 year old technology. It does have its place in a low data setting but their construction requires getting a lot of tedious details right usually as part of a registration pipeline. If you have the data for it I would go for a modern approach like DDPM.
Čestitke Dženan zajedno sa timom!
Research Associate, University of Cambridge
4wI would like to share my perspective, if I may. You have to be very careful when using statistical models in medical imaging, especially for patient diagnosis. I developed a method for 3D reconstruction of the proximal femur from a single 2D radiograph (using ITK and VTK) during my PhD more than a decade ago. It was later commercialized as 3D-DXA (3D-Shaper), and is now even sold by Fujifilm with regulatory approval for patient diagnosis in many countries worldwide. Unfortunately, the software does not work as intended. One major problem is that reconstructing 3D anatomy from a single 2D image can produce results that look plausible but do not actually represent the patient’s true anatomy. Common evaluation metrics such as surface distances or simple correlations are misleading and cannot establish accuracy. I explain these and many other limitations of this approach in my review article in JBMR Plus: https://doi.org/10.1093/jbmrpl/ziaf075 That is not to say there are not many valid applications of statistical shape models and 2D-to-3D methods in medical imaging, but they must be applied with caution and validated appropriately.