Convolutional Neural Network for Anatomical Image Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Manual visual assessment of medical anatomical images is time-consuming and challenging due to the large number of possible features and combinations, with human interpreters often missing findings or providing subjective clinical interpretations.
Innovation Solution
A computer-implemented method using convolutional neural networks (CNNs) to identify visual findings in anatomical images by processing pairs of images captured at unique orientations, with feature vectors concatenated and analyzed through a fully connected layer, trained on datasets generated from text-based radiology reports to automate the detection of multiple possible findings.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual visual assessment by radiologists is used to identify visual findings in anatomical images, then clinical interpretation and diagnosis can be provided, but the process is time-consuming and accuracy is reduced due to the large number of possible features and combinations that must be processed
Solution Approach 1:
The patent replaces the mechanical system of manual radiologist assessment with an automated computer-based system using deep learning models. The system processes anatomical images through trained neural networks that automatically identify visual findings, substituting human visual inspection with computational analysis to improve accuracy and reduce time loss.
Solution Approach 2:
The patent creates a digital copy of the radiologist's diagnostic process through trained deep learning models. These models are trained on large datasets of annotated images to replicate and enhance human diagnostic capabilities, allowing the system to identify visual findings with high accuracy without the time constraints of manual review.
2Reliability
If the statistical classifier processes multiple image orientations and combinations of findings, then the ability to detect visual findings improves, but the device complexity increases
Solution Approach 1:
The patent segments the complex classification task into multiple specialized components: separate deep learning models are trained for different image orientations (PA and lateral views), and the system processes each orientation independently before combining results. This segmentation allows the system to handle multiple findings reliably while managing complexity through modular architecture.
Solution Approach 2:
The patent creates a universal classification system that handles multiple visual findings across different image orientations using the same deep learning framework. The trained models are designed to identify various types of visual findings (consolidation, atelectasis, pneumothorax, etc.) in both PA and lateral views, providing multi-functional capability without requiring separate specialized systems for each finding type.
Data Source
AI summary
There is provided a method comprising: providing two anatomical images of a target individual, each captured at a unique orientation of the target individual, inputting first and second anatomical images respectively into a first and second convolutional neural network (CNN) of a classifier to respectively output first and second feature vectors, inputting a concatenation of the first and second feature vectors into a fully connected layer of the classifier, and computing an indication of distinct visual finding(s) present in the anatomical images by the fully connected layer, wherein the statistical classifier is trained on a training dataset including two anatomical images of each respective sample individual, each image captured at a respective unique orientation of the target individual, and a tag created based on an analysis that maps respective individual sentences of a text based radiology report to one of multiple indications of visual findings.