Autoencoder Anomaly Detection for Medical Imaging Detectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Medical imaging systems, such as PET scanners, face challenges in detecting defects and anomalies in their hardware and software modules without prior knowledge of the defects, which can impair system performance and require manual or empirical inspection.
Innovation Solution
An autoencoder-based method is employed to acquire data from detectors, learn normal patterns, and identify anomalies by reconstructing input data, allowing for automated detection of defects and performance issues in medical imaging systems without requiring a priori knowledge of the defects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional empirical algorithms and visual inspection are used to detect detector defects, then the detection process requires a priori knowledge of defect types and manual interpretation, but this approach cannot automatically identify novel or unexpected anomalies
Solution Approach 1:
The autoencoder performs self-service by automatically learning the normal operating patterns of the medical imaging system from training data and using this learned knowledge to autonomously detect anomalies without requiring manual programming of defect types or human interpretation. The system serves itself by encoding normal patterns and automatically flagging deviations, eliminating the need for external expert knowledge for each new defect type.
Solution Approach 2:
The system transforms the detection approach by changing from fixed empirical thresholds to dynamic learned representations. The autoencoder learns optimal parameter configurations and decision boundaries from data, allowing the system to adapt to new defect types automatically. The reconstruction error metric dynamically adjusts to what constitutes an anomaly based on learned normal variations.
2Reliability
If manual visual inspection of detector maps is performed, then interpretation can be performed by human experts, but the process is time-consuming and cannot operate continuously
Solution Approach 1:
The patent replaces the mechanical human inspection process with an automated computational system. The autoencoder neural network substitutes human visual inspection and interpretation with machine learning-based pattern recognition. This substitution enables continuous automated operation while maintaining or improving detection reliability through consistent application of learned patterns without human fatigue or variability.
3Ease of manufacture
If empirical algorithms are used to interpret detector characteristics, then the interpretation can be performed with existing knowledge, but the algorithms cannot detect anomalies outside predefined patterns
Solution Approach 1:
The system performs preliminary action by training the autoencoder on normal operating data before actual anomaly detection begins. This pre-learning phase establishes the baseline of normal patterns that the system will use to identify deviations. By preparing the model in advance with comprehensive training, the system gains the ability to detect previously unseen anomalies while maintaining implementation simplicity through a standardized training-deployment workflow.
Data Source
AI summary
A method for detecting an anomaly related to a medical imaging device includes acquiring data from a plurality of detectors of the medical imaging device, applying the acquired data to a first autoencoder, and detecting, based on outputs from the first autoencoder, an anomaly related to the medical imaging device.