AI Cell Classification With Selective 3D Imaging for Lung Cancer
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current lung cancer detection methods have poor sensitivity and specificity, leading to missed diagnoses and increased costs due to invasive procedures, and existing optical tomography systems generate unnecessary 3D images of cells that are not indicative of lung cancer.
Innovation Solution
An AI-based cell classification method that generates a representative 2D image of cells using optical tomography, analyzes these images with 2D classifiers to determine if a 3D image is needed, and only generates 3D images for cells with abnormal or BEC-like features, thereby optimizing resource use and reducing processing time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If 3D images are generated for all cells in the patient sample, then complete analysis coverage is achieved, but processing time and resource consumption increase significantly
Solution Approach 1:
The system performs preliminary 2D classification of cells before generating 3D images. The 2D classifier evaluates each cell and identifies only those with abnormal features or BEC-like features as candidates for 3D imaging, allowing the system to prepare and prioritize 3D image generation for relevant cells only, thus reducing overall processing time while maintaining analysis coverage for diagnostic cells
Solution Approach 2:
The system extracts and isolates only the subset of cells that require 3D imaging based on 2D classification results. By separating the cell population into two groups (those needing 3D analysis and those that don't), the system applies 3D imaging selectively to relevant cells, reducing the total number of 3D images generated while ensuring complete analysis of diagnostic cells
2Reliability
If 3D images are generated for all cells, then no diagnostic cells are missed, but computational resources and storage are wasted on non-diagnostic cells
Solution Approach 1:
The system performs preliminary 2D classification to identify diagnostic cells before applying computationally intensive 3D imaging. This preliminary screening ensures that only cells with potential diagnostic value (abnormal or BEC-like features) undergo 3D analysis, preventing waste of computational resources on non-diagnostic cells while maintaining reliability by ensuring all potential diagnostic cells are captured
Solution Approach 2:
The system applies different quality levels of analysis to different cells based on their diagnostic potential. Cells identified as abnormal or BEC-like receive high-quality 3D imaging, while other cells receive only 2D analysis. This differentiated approach optimizes resource allocation by matching computational effort to diagnostic need, reducing overall resource consumption while maintaining diagnostic accuracy
3Productivity
If 2D classification is performed before 3D imaging, then processing efficiency improves, but risk of misclassification increases
Solution Approach 1:
The classification process is segmented into two distinct stages: 2D classification for initial screening and 3D classification for definitive diagnosis. The 2D classifier rapidly identifies potential diagnostic cells, and the 3D classifier then performs detailed analysis on this subset. This segmentation allows the system to leverage the speed of 2D analysis while using 3D analysis to confirm and refine classifications, thereby maintaining high accuracy while improving overall efficiency
Solution Approach 2:
The system implements feedback between 2D and 3D classification stages. Results from 2D classification guide which cells undergo 3D imaging, and 3D classification results can feedback to refine 2D classification thresholds and parameters. This feedback mechanism allows the system to learn from actual diagnostic cases, continuously improving classification accuracy while maintaining the efficiency benefits of the two-stage approach
Data Source
AI summary
The present disclosure provides a system and method for AI-based cell classification of cells from a patient sample to determine if cells indicative of lung cancer are present. In the system and method, 2D imaging is used to eliminate cells not likely to be indicative of lung cancer from subsequent 3D imaging, while 3D imaging is conducted for cells likely to be indicative of lung cancer. The present disclosure further provides a method of training 2D cell classifiers for use in the system and method for AI-based cell classification.