Virtual Patient 3D Mesh Generation With Plausibility Constraints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing virtual patient models for medical training are limited due to restricted access to complete patient datasets, leading to reduced training scenarios, high development costs, and compliance with data privacy regulations, which hinders the creation of a diverse range of models necessary for effective physician training and artificial intelligence training.
Innovation Solution
A system and method for generating a virtual patient model using a morphable 3D mesh template with adjustable anatomical parameters, plausibility relations, and a pseudo-random number generator to create diverse and plausible virtual models, supported by cloud computing for data storage and simulation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If virtual patient models are created from real patient datasets, then the realism and accuracy of the models are improved, but the number of available models is limited and development costs increase
Solution Approach 1:
The patent creates virtual copies of patient data by training an artificial neural network on real patient datasets. The trained model can then generate synthetic patient data that replicates the statistical properties and anatomical variations of real patients without requiring actual patient records, enabling unlimited model generation at low cost
Solution Approach 2:
The system varies anatomical and physiological parameters within realistic ranges derived from training data to generate diverse virtual patient models. By systematically changing parameters such as organ dimensions, tissue properties, and physiological values, the system creates numerous unique models while maintaining medical plausibility
2Adaptability or versatility
If more virtual patient models are generated to increase training scenarios, then the versatility of training is improved, but data privacy regulations and patient consent limitations worsen availability
Solution Approach 1:
The artificial neural network learns the underlying distributions and relationships in real patient data during training, then generates synthetic copies that preserve these characteristics. This allows unlimited generation of diverse training scenarios without accessing additional real patient records, fully complying with data privacy regulations
Solution Approach 2:
A single trained neural network model can generate diverse virtual patients across multiple organ systems and pathological conditions, making the system universally applicable for various training scenarios including cardiology, neurology, oncology, and more, all from one data source
3Manufacturing precision
If virtual models are created one by one by specialized experts, then the quality and plausibility of models are improved, but the time and cost of development increase
Solution Approach 1:
The artificial neural network automatically generates plausible virtual patient models without requiring manual intervention from experts for each model. The system self-regulates by learning from training data what constitutes realistic patient anatomy and physiology, then autonomously generates new models that meet quality standards
Solution Approach 2:
The trained neural network encapsulates expert knowledge about patient anatomy and physiology learned during training, then automatically reproduces this expertise when generating new models. This transfers expert-level model creation capability from manual processes to automated systems
Data Source
AI summary
A system for generating a virtual model of a virtual patient including: a template module to provide a morphable virtual 3D mesh template for a virtual model of a human organ; an anatomical parameters module to provide a range of values for each adjustable anatomical parameters of the virtual model, each value associated with a morphed state of the template; a plausibility module to provide at least one relation for the template in at least one morphed state, the relation being between at least two of the parameters; an instantiation module to provide, using a pseudo-random number generator, a parameter set including, for each parameter of the virtual model, a value of the corresponding range of values, while fulfilling the at least one relation; and a generating module to generate the virtual model based on the template in the morphed state corresponding to the selected parameters.