Surgical Microscopy Data Classifier for Training Set Balancing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Acquiring large amounts of high-quality data for training machine learning instruments in the medical field is challenging due to the vast amount of high-resolution data generated by surgical microscopy systems, which is difficult to transmit and store, and existing methods do not effectively balance the representation of frequent and rare medical interventions.
Innovation Solution
A method where a classifier analyzes new data sets from surgical microscopy systems to determine their significance for training machine learning instruments, deciding which data sets to store in a central database based on their contribution, using automated image analysis and additional information, and updating the classifier periodically to maintain a balanced data set ratio.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If all data from surgical microscopy systems are stored in a central database for training machine learning instruments, then the quantity and quality of training data increases, but the network bandwidth requirement and data transmission complexity increases excessively
Solution Approach 1:
The patent applies preliminary action by performing data filtering and classification at the source (surgical microscopy system) before transmission to the central database. The system pre-determines which data should be transmitted based on predefined criteria, thereby reducing the overall data volume and transmission complexity while ensuring high-quality training data is collected.
Solution Approach 2:
The patent extracts only the essential and high-value data from the total data stream generated by surgical microscopy systems. By identifying and extracting only the most relevant data for training machine learning instruments, the system reduces transmission bandwidth requirements while maintaining data quality for effective model training.
2Measurement precision
If high-resolution data from surgical microscopy systems is transmitted to a central database, then the quality of training data improves, but the network capacity and transmission time increase excessively
Solution Approach 1:
The system performs preliminary filtering and quality assessment of data before transmission. By pre-evaluating data quality and selecting only the most valuable high-resolution data for transmission, the system reduces transmission time while ensuring that only high-quality data is sent to the central database for training.
Solution Approach 2:
The patent extracts only the essential high-quality data from the total data stream, removing redundant or low-value data. This extraction process reduces the amount of data that needs to be transmitted, thereby reducing transmission time while maintaining the quality necessary for effective machine learning training.
3Quantity of substance
If data from frequent medical interventions are stored in large quantities, then the volume of training data increases, but the representation balance between frequent and rare interventions deteriorates
Solution Approach 1:
The patent applies local quality by differentiating the treatment of different data types based on their specific characteristics. The system assigns different weighting and storage priorities to data from frequent versus rare interventions, ensuring that rare interventions receive appropriate representation in the training data while still accommodating the volume needed from frequent procedures.
Solution Approach 2:
The system changes the parameter of data selection criteria to include not only data volume but also intervention rarity as a factor. By modifying the selection parameters to consider both quantity and representation balance, the system ensures that the training data reflects the full spectrum of medical interventions appropriately, preventing over-representation of common procedures.
Data Source
AI summary
A method for acquiring data with the aid of a surgical microscopy system comprises recording a data set having a multiplicity of images and deciding whether or not the new data set ought to be stored in a data memory of a database. The decision is taken by an existing classifier. A new classifier is determined on the basis of training data which comprise the data sets stored in the data memory of the database and/or comprise data obtained from the data sets stored in the data memory of the database. The new classifier is then used instead of the existing classifier when deciding whether a subsequently recorded new data set ought to be stored.