Building and training a Predictive Analysis Model that can read Pre- and Post-operative scans (X-rays, CAT, MRIs) to predict the outcomes and success probability for proposed Spinal implants.
Business Vertical:
Healthcare & Life Sciences
Region:
USA
Size of the firm:
Small to Medium
Challenge:
Improving the prediction accuracy of spinal surgery outcomes amid varied patient history, lack of documentation, and non-standard scans
Client
Client is a developer, manufacturer and marketer of specialized, proprietary medical devices serving the orthopedic and neurosurgery communities in over 30 countries across the globe. They are considered an innovator in spinal pathology sector.
Challenge
The surgical treatments of spine-related conditions often require the placement of pedicle screws and implants to offer better mechanical stability to the patient. A mal-positioned implant can lead to severe neurological ramifications. Consultation based on manually segmented vertebral models is time consuming and often error-prone. Initial attempts to automate segmentation by the client were unsuccessful due to the complexity of the spinal structure, vertebrae changing in shape with each movement, and lack of qualitative testing data.
Proposed Solution
- Build and train a Vertebrae Segmentation Model that can automatically localize and segment the 3D vertebral bodies in patient’s anatomy.
- Create a Pre-op and Post-op Image Discrimination and Calibration Model that can identify and analyze shapes, vertices while accurately calculating the distances and angles based on coordinates.
- Evaluating the model with data from clinical workflow.
What Qentelli did
Qentelli commenced the project in a phased program:
The project started with our team conducting workshops with the stakeholders to identify the use cases and prioritize them based on the complexity and business readiness. Our team of AI-ML experts drafted multiple algorithms and evaluated them for accuracy and quality of information. The finalized segmentation methodology consisted of two pipelines: Region-Based Segmentation and Shape-Prior Based Flows for Individual Vertebra Segmentation.
Qentelli‘s team of researchers and data scientists collected scans from hundreds of patients, including X-Ray, CAT and MRI, and associated additional information on medical history, such as age, gender, underlying health condition, medications if any, post-op health information and overall clinical outcomes along with the scans. The data was well-sanitized to remove any personal information to ensure patient privacy and compliance with HIPAA and other applicable laws.
We built a patent-pending Machine Learning Algorithm that can detect and identify Regions Of Interest In Biomedical Images Using Deep Neural Networks. The architecture consists of 3 models in total – Object Identification Model, Instance Segmentation Model, and Customer Network (Regression) Model. We have used a state-of-the-art version of Faster Region Based Convolutional Neural Networks (R-CNN) to execute the proof of concept.
.svg)
Faster R-CNN method used for the Object identification
Scans from hundreds of patients (X-Ray, CAT and MRI) were collected from client’s database, classified and selected based on Scan quality, Additional information on medical history (sanitized data) was also collected and associated with the scans.
Computer Vision and ML Algorithms are used to detect Spinal implants as well as understand “positive corrections” (comparing before and after surgery). Accuracy scores were assigned to the algorithm efficiency. Given a medical record with pre-surgery scans, the Algorithm can identify possible implant locations and surgical outcomes for Surgeons to review.
The prediction model was trained using a common model training strategy and removed leaking features. After infusing a pre-trained NLP, the program is deployed as a web application that Surgeons can use to augment their decision process – the app would provide recommendations for implant locations and surgical outcomes given a PHR/EHR and pre-surgery scans. The surgeons can then review the predictions and make informed decisions about the implant's placement and expected outcomes.
“The novel automatic program designed and built by Qentelli has obviated the need for time-consuming manual segmentations and eliminated errors due to inter- and intra-observer variability.”
- Dave, President
Outcomes
We tested the model and the app in a pilot study with 25 hospitals and 100 surgeons, and the results were extremely positive. Through observation of the users who were not very tech-savvy, we found ways to improve the way the software worked by aligning closely with the workflows used by the clinical specialists.
The hospitals and surgeons were also delighted by AI assisting them in delivering much better outcomes – a testament to the power of AI and how it can be ethically used to improve human life!
- Improved accuracy of diagnoses: To validate the combinational model performance, the team used a dataset of pre- and post-surgery scans from over 4500 patients. The algorithm was able to accurately predict the implant's location and type in 87% of cases and predict the surgical outcomes in 92% of cases.
- Reduced risk of complications: Our initial research shows that the machine learning algorithm was able to identify patients who were at risk for developing complications after spinal fusion surgery. This information could then be fed to help surgeons plan accordingly and reduce the risk of complications.
- Improved patient satisfaction: Our research found that patients who had their spinal implants surgery planned using a machine learning algorithm were more satisfied with their care than patients who had their surgery planned by a human doctor. This suggests that patients may be more satisfied with their care when it is planned using a machine learning algorithm.
Conclusion
The use of AI-ML algorithms in predicting spinal implant outcomes can provide a valuable addition to the toolkit for surgeons, hospitals, and other providers. The success of this case study shows not only the potential of AI-ML in the healthcare industry, highlighting the importance of collecting and analyzing large amounts of data to develop accurate and effective algorithms, but more importantly, the right use of technology in preserving privacy and ethics.
Let's discuss your business challenge next
Get in touch with our experts for simple and sustainable solutions to your business transformation challenges.
%20(1).png){kind=small}
%20(1).jpg){kind=icon}