AI-Powered MedTech Breakthrough: CSIRO and Singular Health Unveil Revolutionary Spinal Vertebrae Segmentation Technology

A groundbreaking AI-powered medical technology, developed through collaboration between CSIRO‘s Data61 and Australian Medical Imaging Company Singular Health, swiftly segments spinal vertebrae with an impressive 95% accuracy rate within a mere two minutes. This innovation holds the promise of revolutionising surgical planning and facilitating the design of customised implants for medical professionals.

Traditionally, the segmentation of spinal vertebrae in computerised tomography (CT) scans has demanded extensive manual labour, involving countless hours of meticulous identification and markups. However, the advent of AI automation heralds a transformative shift in this arduous process, significantly reducing time and effort while ensuring exceptional segmentation precision and localisation accuracy, as elucidated by Dr. Dadong Wang, Research Lead at Data61.

Singular Health’s Executive Director of Innovation, Dr. Guan Tay, underscores the game-changing potential of this automated segmentation technology. By integrating AI-driven automation into the segmentation process, medical professionals will now only need to make minor adjustments and validate the software’s outputs. This semi-automated approach empowers surgeons and radiologists to fine-tune the results according to their interpretations, ensuring meticulous compliance with image analysis standards while substantially streamlining processing time.

Advertisements

The utilisation of artificial intelligence in medical imaging, particularly in radiology, stands poised to profoundly reshape workflow dynamics for radiologists.

Leveraging a comprehensive dataset comprising over 200 CT scans of labelled data, the Data61 team meticulously explored various AI models and pre-processing techniques to achieve precise instance segmentation, labelling, surface meshing, and spatial localisation of individual vertebrae.

Dr. Wang elaborates on the AI development process, highlighting the adaptation of deep learning-based instance segmentation methodologies such as nnUNET, SC-NET, and Dense-NET. These models were rigorously trained using the VerSe’2020 dataset, comprising 100 CT scans of spines from individuals spanning diverse age groups and genders. Subsequently, the trained models underwent rigorous testing on an additional 100 CT scans, generating segmented labels of the spine, individual vertebrae, spatial coordinates, and vertebra identification.

The integration of this cutting-edge technology into Singular Health’s MedVR software represents a significant milestone, offering a transformative solution for hospitals, clinicians, educational institutions, and universities alike. This milestone achievement was made possible through the CSIRO Kick–Start initiative, which extends funding and support to innovative Australian start-ups and small businesses, granting access to CSIRO’s unparalleled research and development (R&D) expertise and capabilities.

For more health news, click here.