Methods for improved surgical planning using machine learning and devices thereof
Inventors
MCGUAN, Shawn P. • Laster, Scott Kennedy • WILKINSON, Zachary C. • Janna, Sied W. • Farley, Daniel
Assignees
Smith and Nephew Orthopaedics AG • Smith and Nephew Asia Pacific Pte Ltd • Smith and Nephew Inc
Publication Number
US-12343085-B2
Publication Date
2025-07-01
Expiration Date
2040-02-04
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Abstract
Methods, non-transitory computer readable media, and surgical computing devices are illustrated that improve surgical planning using machine learning. With this technology, a machine learning model is trained based on historical case log data sets associated with patients that have undergone a surgical procedure. The machine learning model is applied to current patient data for a current patient to generate a predictor equation. The current patient data comprises anatomy data for an anatomy of the current patient. The predictor equation is optimized to generate a size, position, and orientation of an implant, and resections required to achieve the position and orientation of the implant with respect to the anatomy of the current patient, as part of a surgical plan for the current patient. The machine learning model is updated based on the current patient data and current outcome.
Core Innovation
The invention provides methods, non-transitory computer readable media, and surgical computing devices that improve surgical planning by employing machine learning techniques. A machine learning model is trained on historical case log data sets comprising patient data, implant data, and healthcare professional data correlated with surgical outcomes. This learned model is applied to current patient data, including anatomy data, to generate a predictor equation. The predictor equation is optimized to recommend implant size, position, and orientation, as well as the one or more bone resection parameters necessary to achieve the implant placement in relation to the patient's anatomy. The surgical plan for the current patient is thus generated and can be controlled intraoperatively, for example, to direct surgical tool actuation.
The technical problem addressed is the need for improved surgical planning and workflow during arthroplasty procedures (such as total knee, partial knee, or total hip arthroplasty) where traditional surgical plans, often developed before surgery, fail to adapt optimally to patient-specific factors encountered intraoperatively. Surgeons typically adjust plans during surgery based on patient anatomy and other real-time information, but existing systems lack dynamic, data-driven capabilities to integrate historical case data and use machine learning to provide optimized implant placement and bone resection parameters efficiently. Additionally, conventional optimization techniques are computationally intensive and lack transparent, real-time recommendations correlating surgical parameters to post-operative outcomes.
The invention solves these issues by training a neural network-based machine learning model on extensive historical surgical data, including outcomes and patient and implant characteristics, and applying it intraoperatively or preoperatively to generate predictive equations. These equations link implant parameters to estimated patient-specific responses and are optimized with techniques such as Monte Carlo sampling or parallel tempering to recommend implant size, placement, orientation, and necessary resections. The system updates the model based on current patient data and observed outcomes, allowing continuous learning. Moreover, the optimized surgical plan can be implemented via robotic or computer-assisted systems, improving the accuracy and personalization of arthroplasty procedures.
Claims Coverage
The claims disclose independent inventive features covering a machine learning-based surgical planning method, a surgical computing device for such planning, and non-transitory computer readable media with instructions to perform the method. These features detail the use and training of neural networks with historical and current patient data to generate and optimize predictor equations for implant positioning and resection parameters, and integrating the surgical plan with control of surgical tools.
Machine learning based surgical planning method
Training a neural network using historical case log data sets that correlate patient, implant, and healthcare professional data with surgical outcomes; applying the trained network to current patient anatomical data to generate a predictor equation linking implant size, position, and orientation to estimated anatomical responses; optimizing the predictor equation to determine implant parameters and corresponding bone resection parameters as part of a surgical plan; controlling surgical tool actuation to perform resections according to the surgical plan; updating the neural network based on current patient data and surgical outcome data.
Neural network architecture and training features
Utilizing neural networks comprising multiple input nodes and downstream nodes connected by weighted connections where weighting values serve as predictor equation coefficients; obtaining a sensitivity threshold to disregard certain input nodes; providing seeding data from historical case logs as network input and iteratively altering weights to match known outcome data.
Surgical computing device for improved surgical planning
Including memory storing programmed instructions and processors that train neural networks based on historical surgical data; apply the networks to current patient anatomical data to generate predictor equations relating implant parameters to anatomical responses; optimize these equations to generate implant parameters and resection parameters forming a surgical plan; control surgical tool actuation based on the plan; optionally update the neural network with new patient and outcome data; employ optimization techniques such as Monte Carlo sampling; generate anatomical data from medical imaging and instruct patient-specific instrumentation systems accordingly; generate intra-operative algorithms with recommended actions, evaluate action results, and update subsequent recommendations considering deviations.
Non-transitory computer readable medium with executable instructions
Executable code causing processors to perform training of neural networks on historical data sets, apply networks to current patient anatomy to generate predictor equations relating implant size, position, and orientation to anatomical responses, optimize the equations to produce implant and resection parameters for surgical plans, control surgical tool actuation consistent with the plan, and update the neural network from current patient data and outcomes; includes features of weighting values as predictor coefficients, applying sensitivity thresholds, providing seeding data from history, and generating intraoperative algorithms with iterative updates based on execution and deviations.
The claimed inventions collectively provide a comprehensive system and method for improved surgical planning employing machine learning models trained on historical data to generate optimized, patient-specific surgical plans. These plans include implant sizing, positioning, orientation, and bone resections, with integration of surgical tool control and adaptive intraoperative plan updates, applicable primarily to orthopedic procedures such as arthroplasty.
Stated Advantages
Provides improved surgical planning using historical data and machine learning to generate patient-specific implant positioning and resection parameters.
Enables intraoperative optimization and dynamic updating of surgical plans based on real-time patient data and surgical outcomes.
Integrates surgical plan execution with robotic or computer-assisted tool control for precise implementation.
Employs neural network models that allow reduction in computational intensity compared to conventional models, facilitating near real-time recommendations.
Documented Applications
Orthopedic surgical planning and execution, especially for total knee arthroplasty (TKA), partial knee arthroplasty (PKA), and total hip arthroplasty (THA).
Use with surgical computing devices to aid in preoperative and intraoperative implant sizing, positioning, orientation, and bone resection planning.
Integration with robotic-assisted surgical systems and patient-specific knee instrumentation (PSKI) systems to implement surgical plans.
Postoperative plan updates and improving implant design and sizing based on accumulated episode of care data.
Interested in licensing this patent?