Image Classification Training Using Multi-Modality Surgical Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image classification techniques in medical imaging, particularly for identifying abnormal tissue portions during surgeries, face challenges in accuracy and reliability, especially for conditions like high-grade gliomas, where even trained physicians may misclassify tissues due to limited information from white light images alone.
Innovation Solution
A computer-implemented method for training an image classification machine-learning model using a dataset of labeled image pairs, including intra-op white light surgery images and images of different modalities (such as blue light or MRI) to improve tissue classification, employing convolutional neural networks for enhanced feature recognition and segmentation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If only white light surgery images are used for training, then the training process is simple and fast, but the classification accuracy of abnormal tissue portions is insufficient
Solution Approach 1:
The patent combines multiple imaging modalities (white light images, blue light images, and MRI images) into a unified training dataset. This merging of different data sources provides the machine learning model with complementary information from various imaging techniques, thereby improving classification accuracy while managing the complexity through systematic data integration.
Solution Approach 2:
The training dataset is constructed as a composite of multiple imaging modalities, analogous to composite materials. Each modality contributes unique characteristics (white light provides anatomical structure, blue light provides fluorescence information, MRI provides detailed tissue characterization), and their combination creates a more robust training foundation for accurate tissue classification.
2Reliability
If multiple imaging modalities are integrated for training, then the classification accuracy of abnormal tissue portions improves, but the training process becomes more complex and time-consuming
Solution Approach 1:
The patent performs preliminary actions by pre-processing and organizing multiple imaging modalities before the actual model training. The training dataset is prepared in advance with aligned and registered images from different modalities, which reduces the computational burden during the training phase and minimizes training time while maintaining high reliability.
3Measurement precision
If multiple imaging modalities are used, then the ability to identify challenging conditions like high-grade gliomas improves, but the device and data processing complexity increases
Solution Approach 1:
The patent segments the imaging system into distinct modalities (white light imaging subsystem, blue light imaging subsystem, and MRI subsystem), each optimized for specific functions. This segmentation allows each component to be independently managed and processed, reducing overall system complexity while maintaining the ability to identify challenging conditions through their combined use.
Data Source
Figure 1
Figure 2A~2B
Figure 3~4
AI summary
A computer-implemented method (100) for training an image classification machine-learning model (108) is provided for identifying abnormal tissue portions in intra-op surgical images. The method may comprise the step of providing a training dataset. The training dataset may comprise a plurality of labelled training images, in particular image pairs, further comprising: an intra-op white light surgery image (102a); an intra-op blue light surgery image (104), the white light surgery image (102a) and the blue light surgery image (104) having matching fields of view; and at least an intra-op white light surgery image (102a); at least one other image of different-modality; and a label (106) for classifying at least one of an abnormal tissue portion and/or another tissue portion in at least one of the white light surgery image (102a) and/or the at least one other image of different-modality. Training the machine-learning model (108) may be based at least in part on the training dataset, thereby obtaining a trained machine-learning model (110) which is configured to identify one or more abnormal tissue portions in an intra-op white light surgery image (102b) captured during a surgery.