Deep Learning Data Rescue for Corrupted Emission Tomography
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Emission tomography imaging often results in poor-quality data due to patient movement or incomplete scans, leading to corrupted data that fails to meet clinical guidelines, which can render images non-diagnostically useful.
Innovation Solution
A machine-learned model is employed to recover information from nuclear imaging data, using deep learning techniques to generate diagnostically useful images even from data that is insufficient or corrupted, such as by filling in missing data or correcting artifacts, and selecting the best model for the specific situation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a scan is performed to acquire emission data, then imaging information is obtained, but patient movement or panic may cause data corruption that fails to meet clinical guidelines
Solution Approach 1:
The patent applies deep learning models to convert corrupted or incomplete scan data into diagnostically useful images. The machine learning system learns from training data to reconstruct high-quality images even when input data is degraded by patient movement, panic, or incomplete scanning, effectively converting harmful data corruption into beneficial diagnostic information
Solution Approach 2:
The patent performs preliminary data recovery and reconstruction using deep learning models before clinical interpretation. By pre-processing corrupted data through trained neural networks that have learned from extensive training data, the system prepares clean, diagnostic-quality images in advance, ensuring reliable results even when scans are compromised
2Loss of information
If scan data is corrupted or incomplete, then diagnostic information is lost, but repeating the scan increases patient burden and time consumption
Solution Approach 1:
The deep learning system converts previously worthless corrupted data into diagnostically useful information, eliminating the need to discard compromised scans and repeat them, thereby recovering diagnostic information without additional time cost
Solution Approach 2:
The patent changes the processing approach by applying deep learning transformations to corrupted data, converting it from an unusable state into a diagnostically valuable state through intelligent reconstruction algorithms that preserve and recover essential diagnostic information
3Measurement precision
If deep learning models are used to recover information from corrupted data, then image quality is improved, but computational complexity increases
Solution Approach 1:
The patent performs computationally intensive deep learning model training and optimization in advance during system setup or offline periods. The trained models are then deployed for rapid inference during actual scanning, separating the heavy computational burden from real-time operation and reducing complexity during clinical use
Solution Approach 2:
The system dynamically adapts by selecting appropriate deep learning models based on the type and severity of data corruption detected. Different trained models can be applied depending on whether the corruption is due to motion, incomplete scanning, or other factors, optimizing computational resource usage while maintaining image quality
Data Source
AI summary
An emission image is generated from poor quality emission data. A machine-learned model may be used to recover information. Emission imaging may be provided due to the recovery in a way that at least some diagnostically useful information is made available despite corruption that would otherwise result in less diagnostically useful information or no image at all.