Microscope Autofocus Using CNN Pixel Difference Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Automated microscopes struggle to match the focusing performance of skilled human operators, especially when dealing with different sample preparation methods and coherent light sources, as existing auto-focus techniques are limited in adaptability and accuracy.
Innovation Solution
A method and system for autofocusing a microscopic imaging system that involves capturing images at multiple focus positions, identifying differences in pixel values, and using historic data to determine the optimal focus position, facilitated by a processor and memory system that trains a machine learning model to predict the best focus position in real-time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If a deep learning model is trained to predict defocus distance using a single image, then the focusing speed is improved, but the adaptability to different sample preparation methods and staining protocols deteriorates
Solution Approach 1:
The patent applies universality by training the deep learning model on a diverse dataset encompassing multiple sample preparation methods, staining protocols, and tissue types. This enables the single model to generalize across different conditions rather than requiring separate models for each protocol, thus achieving both speed and adaptability
Solution Approach 2:
The patent employs preliminary action by pre-training the model on extensive historical data from various sample types and preparation methods before deployment. This pre-training establishes a robust foundation that allows the model to adapt quickly to new samples without retraining, maintaining both speed and versatility
2Measurement precision
If multiple images at different focus positions are captured and processed, then the focusing accuracy is improved, but the time required for focusing increases
Solution Approach 1:
The patent applies skipping by using the deep learning model to predict the optimal focus position directly from a single or minimal number of images, bypassing the need to capture and process multiple images through traditional focus stacking methods. This dramatically reduces focusing time while maintaining accuracy
Solution Approach 2:
The patent replaces the mechanical/iterative process of capturing multiple images and computationally determining focus with a direct neural network prediction. The model learns the complex relationship between image features and focus position, substituting multiple capture steps with a single inference operation
3Measurement precision
If the deep learning model is trained with coherent light sources, then the performance on specific samples is improved, but the generalization to different light sources deteriorates
Solution Approach 1:
The patent applies universality by incorporating data from multiple light sources (coherent and incoherent, different wavelengths) into the training dataset. This enables the model to learn invariant features that generalize across lighting conditions rather than overfitting to coherent light characteristics
Data Source
AI summary
The present disclosure provides method and system for auto focusing a microscopic imaging system using machine learned regression system such as Convolutional Neural Network (CNN). The method comprises receiving a first image of a sample under review and a second image of sample wherein the first image is captured at first focus position and second image is captured at second focus position. The CNN is trained using the plurality of historic difference images along with direction of focus and optimal focus position. The difference of two images are obtained in terms of difference in pixel values. The direction of focus and optimal focus position for difference image is identified based on plurality of historic difference images along with direction of focus and optimal focus position. The method enables automated stage comprising sample to move towards direction of focus and position at optimal focus position for capturing a focused image.