Skin Abnormality Classification via CNN and AR Guidance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current convolutional neural networks (CNNs) for classifying skin abnormalities, particularly skin cancer, have low prediction accuracy on mobile-device camera images due to differences in resolution, quality, and lighting compared to dermatoscopic images, and lack skin-specific boundary mapping and contour detection, which are essential for precise medical interventions.
Innovation Solution
A novel CNN architecture optimized for mobile and dermatoscopic images using proprietary pre-processing, hyperparameter optimization, and biologically inspired weight initialization, combined with a Boundary-Attention Mapper (BAM) system for fine-grained boundary and attention mapping, and integration with augmented reality for spatial measurement and treatment guidance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional CNNs are used for skin abnormality classification, then the system can perform basic classification, but the prediction accuracy is low due to resolution and quality differences between mobile camera images and dermatoscopic images
Solution Approach 1:
The patent applies parameter changes by optimizing CNN hyperparameters specifically for mobile camera images rather than dermatoscopic images. The system adjusts learning rates, batch sizes, and network architecture parameters to accommodate the lower resolution and varying quality of mobile images, thereby improving prediction accuracy without requiring high-resolution input images
Solution Approach 2:
The patent uses transfer learning where a pre-trained CNN model (originally trained on high-quality dermatoscopic images) is copied and adapted for mobile camera images. The pre-trained weights provide a foundation that is then fine-tuned on mobile image datasets, allowing the model to leverage learned features from high-quality images while adapting to the characteristics of lower-quality mobile images
2Manufacturing precision
If conventional CNNs are used, then the system can classify skin abnormalities, but it lacks skin-specific boundary mapping and contour detection capabilities needed for precise medical interventions
Solution Approach 1:
The patent segments the CNN output into multiple functional components: classification outputs, boundary maps, and contour detections. By separating these functions into distinct processing streams, the system can optimize each component independently - using specialized layers for boundary detection while maintaining overall system manageability
Solution Approach 2:
The patent adds spatial dimensionality to the classification output by generating boundary maps and contour detections at multiple resolution levels. This dimensional expansion allows the system to provide both coarse classification and fine-grained boundary information simultaneously, enabling precise surgical guidance without proportionally increasing computational complexity
3Reliability
If high-resolution dermatoscopic images are used for training, then the CNN can achieve good classification performance, but the system cannot accurately measure and segment skin abnormalities in mobile camera images
Solution Approach 1:
The patent changes the input resolution parameters and scaling factors to accommodate mobile camera image qualities. The system dynamically adjusts image resizing operations and measurement scaling based on the actual input image characteristics, allowing accurate measurement of skin abnormalities regardless of the original image resolution
Solution Approach 2:
The patent introduces intermediate representation layers that serve as mediators between the high-resolution training data and low-resolution mobile images. These intermediate layers perform multi-scale feature extraction and size normalization, enabling the system to maintain measurement accuracy across different image qualities by translating between resolution domains
Data Source
AI summary
Systems and methods for use in classifying and measuring images of skin abnormalities. The systems and methods can be used in an AR enabled system that may be used to assist in skin surgeries and skin abnormality triaging and diagnosis. The system uses a convolutional neural network to classify a skin abnormality in an initial image. The CNN may also be used in determining the boundaries of the skin abnormality. A fiducial marker may be present in the initial image and this marker may be used in automatically measuring the size of the skin abnormality. An adjusted image is generated based on the measured abnormality and this adjusted image can be used as an overlay in an AR enabled system for use in assisting surgical procedures.