Imaging System Intermediate Representation for Privacy-Preserving Deep Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing non-invasive imaging technologies face challenges in leveraging deep learning improvements across different systems due to the need for access to raw imaging data, which may contain sensitive patient information, making it difficult to share and train deep learning algorithms effectively.
Innovation Solution
A method where an imaging system generates an intermediate representation of imaging data and transmits it to a central server for training a deep neural network, allowing for the development of global deep learning models without sharing the raw data, thus preserving patient privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If raw imaging data is transmitted to train deep learning models, then model training effectiveness is improved, but patient privacy is compromised
Solution Approach 1:
The patent extracts only the necessary training information from raw imaging data by generating intermediate representations through deep neural networks. These intermediate representations contain the essential features for model training while excluding sensitive patient information, thus resolving the contradiction between training effectiveness and privacy protection.
Solution Approach 2:
The patent introduces intermediate representations as an intermediary between raw imaging data and deep learning model training. This intermediary form allows training information to be transmitted without exposing raw patient data, effectively mediating between the needs of model training and patient privacy protection.
2Object-affected harmful factors
If intermediate representations are transmitted instead of raw data, then patient privacy is preserved, but training data quality may be reduced
Solution Approach 1:
The patent transforms raw imaging data into intermediate representations by changing the data parameters and format. This transformation process extracts essential training features while removing sensitive information, maintaining training quality without compromising privacy through parameter transformation rather than data loss.
3Adaptability or versatility
If deep learning models are trained locally at each imaging system, then model customization is improved, but data sharing and collaborative learning are limited
Solution Approach 1:
The patent creates intermediate representations that serve multiple functions: they can be used for local model training at individual imaging systems while also being suitable for centralized aggregation and collaborative learning across multiple systems. This universal format enables both customization and collaboration without requiring raw data sharing.
Data Source
AI summary
Methods and systems are provided for capturing deep learning training data from imaging systems. In one embodiment, a method for an imaging system comprises performing a scan of a subject to acquire imaging data, inputting the imaging data to a deep neural network, displaying an output of the deep neural network and an image reconstructed from the imaging data, and transmitting an intermediate representation of the imaging data generated by the deep neural network to a server for training a central deep neural network. In this way, imaging data may be leveraged for training and developing global deep learning models without transmitting the imaging data itself, thereby preserving patient privacy.