Learning Device for Virtual Fluorescent Area Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In rigid endoscope systems, the need to administer fluorescent substances like indocyanine green to observe lesions is inconvenient, as it requires additional steps and substances.
Innovation Solution
A learning device and medical image processing device that use paired training images captured in different wavelength bands to specify singular areas based on fluorescence intensity or pixel level, allowing for machine learning to generate a learning model that estimates fluorescent areas without the need for fluorescent substances.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If fluorescent substances are administered to observe lesions, then fluorescent areas can be clearly observed, but the procedure becomes more complex and less convenient
Solution Approach 1:
The patent creates a virtual fluorescent area image that copies the appearance and characteristics of actual fluorescent imaging without requiring real fluorescent substances. This virtual copy is generated through machine learning models trained on paired images, allowing the system to replicate fluorescent area detection while avoiding the complexity of substance administration
Solution Approach 2:
The patent replaces the mechanical/biological process of fluorescent substance administration and excitation with an information processing system. Instead of using physical fluorescent agents and light excitation mechanisms, the system uses machine learning algorithms to directly identify and segment fluorescent areas from normal light images, substituting complex physical processes with computational methods
2Measurement precision
If fluorescent substances and near-infrared excitation light systems are used, then fluorescent observation is achieved, but system configuration becomes more complex
Solution Approach 1:
The patent extracts and removes the fluorescent substance administration step and near-infrared excitation light system from the overall imaging process. By training a machine learning model on paired images (normal light and fluorescent), the system separates the fluorescent area detection function from the physical fluorescent imaging apparatus, allowing normal light images alone to suffice for subsequent observations
Solution Approach 2:
The patent makes the normal light imaging system universal by enabling it to perform both normal light observation and fluorescent area detection functions. The machine learning model allows the same imaging system to derive multiple types of information (anatomical structure from normal light, fluorescent areas from learned patterns) without requiring specialized fluorescent imaging hardware
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach improves convenience by eliminating the need for fluorescent substance administration, simplifies the system configuration by removing the need for near-infrared excitation light, and enhances image processing by clearly distinguishing fluorescent areas.
Implementation Method 1
normal light that is white light is emitted to the observation target, and the normal light reflected by the observation target is captured by an imaging element
Implementation Method 2
the excitation light that excites the fluorescent substance such as indocyanine green is emitted to the observation target, and fluorescence from the observation target excited by the excitation light is captured by a high-sensitivity imaging element
Data Source
AI summary
A learning device 3 includes: a training image acquisition unit 321 that acquires training images in which a first training image acquired by capturing light from a subject irradiated with light in a first wavelength band and a second training image acquired by capturing light from the subject irradiated with light in a second wavelength band different from the first wavelength band are paired; a singular area specification unit 322 that specifies a singular area in the second training image; a first feature data extraction unit 323 that extracts feature data of a singular-corresponding area at a pixel position corresponding to the singular area in the first training image; and a singular-corresponding area learning unit 324 that generates a learning model by performing machine learning on the singular-corresponding area based on the feature data.