Deep Learning Autofocus for Microscope Specimen Stage
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional microscopy focusing methods are time-consuming due to the need for sweeping through the Z-axis to estimate the focal plane, which can impede high-throughput production and cause photobleaching of light-sensitive samples.
Innovation Solution
A deep learning model, specifically a regression convolutional neural network (CNN), is used to predict the distance to the focal plane from two images taken at an arbitrary initial position, allowing for automatic adjustment of the specimen stage without the need for additional hardware.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional Z-axis searching algorithm is used to find focal plane, then focusing accuracy is achieved, but focusing time increases significantly
Solution Approach 1:
The patent replaces the mechanical Z-axis searching algorithm with a deep learning-based autofocus model that uses a convolutional neural network to predict focal plane distance directly from image data, eliminating the need for time-consuming mechanical Z-axis scanning while maintaining focusing accuracy
Solution Approach 2:
The patent introduces an intermediate autofocus model that acts as a mediator between image capture and focal plane determination, using a convolutional neural network to process image data and predict focal distance, thereby avoiding direct mechanical searching through the Z-axis range
2Measurement precision
If prolonged exposure to light is applied during traditional focusing, then focal plane can be determined, but photobleaching of light-sensitive samples occurs
Solution Approach 1:
The patent performs preliminary focusing using the deep learning model that requires minimal light exposure, determining the focal plane before any prolonged imaging occurs, thereby preventing photobleaching of light-sensitive samples while still achieving accurate focus
Solution Approach 2:
The patent replaces light-intensive traditional focusing methods with a computational approach using a convolutional neural network that can determine focal plane from standard image data without requiring prolonged or intensified light exposure
3Measurement precision
If contrast scoring metric evaluation is performed across wide Z-axis range, then accurate focal plane is identified, but throughput of specimen inspection decreases
Solution Approach 1:
The patent replaces the mechanical Z-axis scanning and contrast scoring evaluation system with a deep learning autofocus model that processes image data computationally to identify focal plane, dramatically reducing the time required per specimen and increasing inspection throughput while maintaining identification accuracy
Solution Approach 2:
The patent introduces an autofocus model as an intermediary that uses a convolutional neural network to directly predict focal plane distance from image data, eliminating the need for wide Z-axis range scanning and contrast scoring evaluation, thereby speeding up specimen inspection while maintaining focal plane identification accuracy
Data Source
AI summary
A computing system receives, from an image sensor, at least two images of a specimen positioned on a specimen stage of a microscope system. The computing system provides the at least two images to an autofocus model for detecting at least one distances to a focal plane of the specimen. The computing system identifies, via the autofocus model, the at least one distance to the focal plane of the specimen. Based on the identifying, the computing system automatically adjusts a position of the specimen stage with respect to an objective lens of the microscope system.