Sequential CNN for Microaneurysm Detection Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting diabetic retinopathy, particularly in its early stages, face challenges due to the small size of microaneurysms, leading to high false positives and inefficiencies in image classification, as existing algorithms struggle to accurately identify these lesions amidst other retinal features.
Innovation Solution
The use of a sequential convolutional neural network (CNN) approach with two models optimized for sensitivity and specificity, combined with image preprocessing techniques such as normalization and augmentation, to enhance feature detection and reduce false positives, allowing for more accurate classification of microaneurysms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If image cropping with lesion centering is applied to improve lesion detection, then detection accuracy for microaneurysms improves, but the number of false positives increases significantly
Solution Approach 1:
The patent segments the retinal image into multiple regions of interest (ROIs) based on lesion detection, then processes each ROI separately through the CNN. This segmentation allows the model to focus on specific areas containing microaneurysms while reducing the impact of false positives in other regions, thereby improving overall detection accuracy without proportionally increasing false positives.
Solution Approach 2:
The patent applies different processing qualities to different regions of the image. Regions containing detected lesions receive enhanced processing through the CNN, while other regions are handled differently. This local quality approach ensures that microaneurysm detection receives maximum attention and processing power, improving detection accuracy in critical areas without unnecessarily processing the entire image, thus reducing false positives.
2Productivity
If traditional CNN models are used for image classification, then processing speed is maintained, but detection precision for small features like microaneurysms deteriorates
Solution Approach 1:
By segmenting the image into ROIs based on lesion locations, the patent reduces the effective processing area for each CNN inference. This allows traditional CNN models to maintain their processing speed while focusing computational resources on smaller regions containing microaneurysms, thereby improving detection precision without sacrificing productivity.
Solution Approach 2:
The patent performs preliminary lesion detection and ROI identification before applying the CNN classification. This preliminary action filters out regions that do not contain lesions, allowing the CNN to process only relevant areas. As a result, traditional CNN models can maintain their speed while achieving higher precision by avoiding unnecessary processing of lesion-free regions.
3Measurement precision
If invasive blood sugar testing is used for diabetes screening, then diagnostic accuracy is improved, but patient comfort and screening accessibility deteriorate
Solution Approach 1:
The patent introduces retinal imaging and CNN-based analysis as an intermediary method between non-invasive screening and invasive blood sugar testing. This intermediary approach allows for accurate diabetes detection through eye examinations, which are non-invasive and easily performable, thereby improving screening accessibility while maintaining diagnostic accuracy through the sophisticated image analysis system.
Data Source
AI summary
Systems and methods are described for image classification. An example method may comprise receiving an image comprising an object of interest and determining, based on a first convolution neural network, a first classification of the image. The first convolution neural network may be optimized for a first factor. The method may comprise determining, based on a second convolution neural network, a first classification of the image. The second convolution neural network may be optimized for a second factor. The method may comprise determining, based on the first classification and the second classification, a characteristic of the object of interest. The method may comprise providing the characteristic of the object of interest.