Hyperspectral Band Selection for CNN Esophageal Lesion Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current endoscopic methods for diagnosing esophageal cancer, particularly in early stages, are cumbersome, prone to artificial errors, and lack accuracy due to reliance on single macroscopic data and invasive imaging technologies, making early detection challenging.
Innovation Solution
A method utilizing hyperspectral imaging with band selection and convolutional neural networks to analyze esophageal tissue images, enabling automatic detection and classification of esophageal cancer by converting input images to hyperspectral format, performing band selection, and applying principal component analysis to enhance feature detection and alignment, thereby simulating narrow-band imaging.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional endoscopy methods (WLI, NBI, chromoendoscopy) are used for esophageal cancer detection, then the detection process can be performed with existing equipment, but the detection accuracy remains limited and manual judgment is required which leads to artificial errors
Solution Approach 1:
The patent replaces manual mechanical judgment processes with an automated computer vision system. The system uses hyperspectral image processing and convolutional neural networks to automatically detect and classify esophageal cancer lesions, substituting the mechanical/manual operations of traditional endoscopy interpretation with automated computational analysis, thereby eliminating artificial errors while improving detection accuracy
Solution Approach 2:
The patent combines multiple imaging modalities (white light imaging, narrow-band imaging, and magnifying endoscopy) into a unified hyperspectral imaging framework. By integrating these different imaging approaches into a single composite system that processes multiple spectral bands simultaneously, the patent achieves comprehensive lesion detection with improved accuracy without requiring separate manual analysis of each modality
2Measurement precision
If hyperspectral imaging with full band processing is used, then comprehensive feature detection is achieved, but computational burden and processing time increase significantly
Solution Approach 1:
The patent extracts and selects only the most relevant spectral bands from the complete hyperspectral data cube for cancer detection. By identifying and isolating the specific wavelength ranges that provide the most diagnostic information for esophageal cancer, the system maintains high detection accuracy while reducing the volume of data that requires computational processing, thereby decreasing processing time
Solution Approach 2:
The patent segments the hyperspectral processing pipeline into distinct stages: initial full-band feature extraction, band selection based on relevance to cancer detection, and focused processing of selected bands using convolutional neural networks. This segmentation allows the system to benefit from comprehensive initial analysis while avoiding the computational burden of processing all spectral bands at full resolution throughout the entire processing chain
Data Source
AI summary
The present application related to a method for detecting image using hyperspectral imaging by band selection. Firstly, obtaining a hyperspectral imaging information according to a reference image, hereby, obtaining corresponded hyperspectral image from an input image, and obtaining corresponded feature values by band selection for operating Principal components analysis to simplify feature values. Then, obtaining feature images by Convolution kernel, and then positioning an image of an object under detected by a default box and a boundary box from the feature image. By Comparing with the esophageal cancer sample image, the image of the object under detected is classifying to an esophageal cancer image or a non-esophageal cancer image. Thus, detecting an input image from the image capturing device by the convolutional neural network to judge if the input image is the esophageal cancer image for helping the doctor to interpret the image of the object under detected.