The Challenge
An innovative spinout from a prominent research institution faced significant variability in their predictive model for spinal surgery complications. Their proprietary regression model relied on geometric measurements of spinal discs derived from MRI scans. However, these measurements were performed manually, leading to high variability and reduced accuracy in predictions.
The Solution
Accelvision developed a robust solution to address these challenges:
- Advanced Segmentation Models: We trained a computer vision model to accurately and consistently segment spinal discs in MRI scans.
- Automated Geometric Calculations: By translating complex logic from research MATLAB code into Python, we automated the extraction of precise geometric features from the segmented discs.
- Continuous Training Pipeline: To ensure ongoing improvements, we designed a continuous training pipeline that retrains the model using new data, enhancing accuracy and adaptability over time.
Key Outcomes
- Achieved improved prediction accuracy by standardizing the measurement process.
- Reduced manual effort for experts, who now only needed to make minor corrections to the disk masks.
- Delivered an efficient and scalable pipeline, enabling reliable and consistent model inputs.
This project demonstrates Accelvision’s expertise in combining cutting-edge AI with domain-specific knowledge to deliver impactful, time-saving solutions that drive precision and efficiency in healthcare analytics.
