Multi-View Specimen Image Classification for HILN Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated diagnostic systems face challenges in accurately and efficiently determining the presence of hemolysis, icterus, and lipemia in serum or plasma portions of specimens due to label obstruction and computational complexity, leading to subjective human error and unreliable test results.
Innovation Solution
A specimen testing apparatus utilizing a single deep neural network (SDNN) that combines a segmentation convolutional neural network (SCNN) and a classification convolutional neural network (CCNN) to process multi-spectral, multi-exposure image data from multiple viewpoints, effectively overcoming label obstruction and improving interferent detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automated machine-vision inspection is used to detect interferents, then productivity is improved, but measurement precision deteriorates due to label obstruction and computational complexity
Solution Approach 1:
The patent segments the inspection task into two distinct neural network components: SCNN for segmentation (identifying serum/plasma regions and excluding labels) and CCNN for classification (detecting interferents). This segmentation allows each network to specialize, improving overall measurement precision while maintaining automated productivity.
Solution Approach 2:
The patent introduces an intermediary processing step where multiple images from different viewpoints are captured and processed to create a composite view. This intermediary representation eliminates label obstructions before interferent detection, resolving the contradiction between automated processing and detection accuracy.
2Measurement precision
If multiple images from multiple viewpoints are captured, then measurement precision is improved by overcoming label obstruction, but device complexity increases
Solution Approach 1:
The patent uses a single deep neural network (SDNN) that integrates both SCNN and CCNN functionalities. This universal network processes multiple viewpoint images and performs both segmentation and classification tasks, reducing device complexity while maintaining the precision benefits of multi-view imaging.
Solution Approach 2:
The patent merges the segmentation and classification networks into a single integrated SDNN architecture. This combining reduces computational overhead and system complexity while preserving the ability to accurately detect interferents by first segmenting relevant regions from multiple viewpoints.
3Measurement precision
If manual visual inspection is used to evaluate specimen integrity, then measurement precision is maintained through expert judgment, but productivity deteriorates due to labor intensity and human error
Solution Approach 1:
The patent implements an automated system that performs specimen integrity assessment without human intervention. The SDNN independently captures images, segments serum/plasma regions, detects interferents, and classifies specimen quality, eliminating labor-intensive manual inspection while maintaining or improving precision through consistent automated judgment.
Solution Approach 2:
The patent replaces the mechanical human visual inspection process with an automated optical system coupled with deep neural network analysis. This substitution eliminates human error and labor constraints, simultaneously improving productivity while maintaining measurement precision through the intelligent SDNN system.
Data Source
Figure 1~2
Figure 3A~3B
Figure 4A~4B
AI summary
A method of characterizing a serum or plasma portion of a specimen in a specimen container includes capturing a plurality of images of the specimen container from multiple viewpoints, stacking the multiple viewpoint images along a channel dimension into a single stacked input, and processing the stacked input with a single deep convolutional neural network (SDNN). The SDNN includes a segmentation convolutional neural network that receives the stacked input and outputs multiple label maps simultaneously. The SDNN also includes a classification convolutional neural network that processes the multiple label maps and outputs an HILN determination (Hemolysis, Icterus, and/or Lipemia, or Normal) of the serum or plasma portion of the specimen. Quality check modules and testing apparatus configured to carry out the method are also described, as are other aspects.