Multispectral Imaging Training With Artificial Labels for Biomarker Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing medical imaging systems require extensive manual labeling and large datasets for training, which is time-consuming and costly, and rely heavily on expert medical practitioners for identifying biomarkers, leading to potential errors and processing burdens.
Innovation Solution
A medical imaging system is trained using multispectral images with artificial labels and self-supervised learning techniques, employing transformer encoders and convolutional models to reduce the need for labeled images and processing burden, utilizing anatomical masks and statistical calculations to enhance biomarker detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If extensive manual labeling and large datasets are used for training medical imaging systems, then training accuracy and reliability are improved, but time consumption and processing costs increase significantly
Solution Approach 1:
The system employs self-supervised learning techniques where the model trains itself using unsupervised learning algorithms on unlabeled multispectral images. The transformer encoder and convolutional models automatically extract features and identify biomarkers without requiring manual labeling, enabling the system to serve itself in the training process and eliminating the time-consuming manual annotation step
Solution Approach 2:
The patent replaces the mechanical process of manual labeling by experts with automated computational methods. The transformer encoder and convolutional models perform the classification task that would otherwise require human medical practitioners, substituting mechanical human labor with automated digital processing systems
2Measurement precision
If expert medical practitioners are relied upon for identifying biomarkers, then diagnostic accuracy is improved, but processing burden and costs increase
Solution Approach 1:
The system substitutes the cognitive processing of expert medical practitioners with automated deep learning models. The transformer encoder and convolutional models perform the complex task of biomarker identification that would otherwise require human expertise, thereby maintaining diagnostic accuracy while eliminating the processing burden on medical professionals
Solution Approach 2:
The patent creates digital copies of the diagnostic capability through trained neural network models. These models replicate the pattern recognition and classification abilities of expert practitioners, allowing multiple copies to process images simultaneously without the constraints of human working capacity
3Reliability
If large datasets of labeled images are used for training, then model accuracy is improved, but the quantity of resources and processing power required increases
Solution Approach 1:
The system generates its own training data through self-supervised learning mechanisms. The transformer encoder creates artificial labels and supervisory signals automatically from the multispectral images, eliminating the need for external labeled datasets and reducing the quantity of resources required for training
Solution Approach 2:
The patent performs preliminary feature extraction and representation learning through unsupervised pre-training on large volumes of unlabeled images. This preliminary action prepares the convolutional models to achieve high accuracy on the target diagnostic task without requiring proportional amounts of labeled data, effectively decoupling training data quantity from model performance
Data Source
AI summary
A multispectral imaging system is trained by the application of an unlabelled multispectral image. Artificial labels are added to each image at corresponding wavelengths of the unlabelled multispectral image in view of training a machine learning system. Spatial and then spatial-spectral features of the multispectral images are extracted in successive training phases. The trained machine learning system may then be used to detect biomarkers or other artefacts in a multispectral image by splitting the multispectral image into distinct wavelength-images, applying masks to the wavelength-images to obtain pixel groups, applying statistical calculation to the pixel groups, assembling statistical calculation results into feature vectors, and using the trained machine learning system to extract, from the feature vectors, positive or negative indications related to the presence of biomarkers of other artefacts in the multispectral image. The machine learning system may be retrained upon processing of each subsequent multispectral image.