Binomial Subsample CNN for SPECT Image Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for analyzing intrinsic flood images in SPECT scanners rely heavily on manual human evaluation, which is subjective, difficult to reproduce, and lacks reliability and accuracy.
Innovation Solution
A binomial subsample data augmented convolutional neural network (CNN) is used to automatically classify SPECT images by generating diverse data sets through bootstrap sampling and geometric transformations, improving classification accuracy and objectivity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual human evaluation is used to analyze intrinsic flood images, then the analysis can be performed with basic equipment, but the reliability and repeatability of the analysis deteriorates due to subjectivity and dependence on engineer experience
Solution Approach 1:
The patent replaces the manual mechanical evaluation process with an automated computer-based CNN system. The convolutional neural network automatically processes intrinsic flood images to detect non-uniformities, replacing the subjective human visual evaluation process and improving reliability while maintaining operational simplicity through software automation
Solution Approach 2:
The CNN system performs self-learning and self-evaluation through automated training and classification. The system uses binomial subsampled data to train the neural network, which then autonomously classifies images without requiring human expert intervention, achieving reliable and repeatable analysis through self-service automation
2Measurement precision
If more training data is used to improve CNN classification accuracy, then the classification precision improves, but the data processing time and computational resources increase
Solution Approach 1:
The patent applies binomial subsampling to create a reduced but representative subset of the training data. Instead of using the complete dataset for training, the method uses a strategically selected subsample that captures the essential characteristics of the data distribution, achieving good classification accuracy with reduced training time and computational resources
Solution Approach 2:
The patent transforms the training data by applying geometric transformations (rotation, flipping, scaling) to the subsampled images. This parameter transformation creates additional training samples from limited data, improving classification accuracy without requiring proportionally more original data or training time
3Adaptability or versatility
If geometric transformations are applied to generate more training images, then the diversity of training data improves, but the processing complexity increases
Solution Approach 1:
The patent applies geometric transformations during the data generation phase before training begins. By pre-processing the training images with rotations, flips, and scaling operations, the system creates a diverse dataset in advance, allowing the CNN to learn from varied perspectives without adding complexity during the actual training and inference phases
Data Source
AI summary
A method for automatically classifying emission tomographic images includes receiving original images and a plurality of class labels designating each original image as belonging to one of a plurality of possible classifications and utilizing a data generator to create generated images based on the original images. The data generator shuffles the original images. The number of generated images is greater than the number of original images. One or more geometric transformations are performed on the generated images. A binomial sub-sampling operation is applied to the transformed images to yield a plurality of sub-sampled images for each original image. A multi-layer convolutional neural network (CNN) is trained using the sub-sampled images and the class labels to classify input images as corresponding to one of the possible classifications. A plurality of weights corresponding to the trained CNN are identified and those weights are used to create a deployable version of the CNN.