Pet Radiology Image Classification With Orientation Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The limited number of veterinary radiologists and the time-consuming nature of image-based diagnostic techniques in pet radiology, coupled with issues like incorrect orientation and missing laterality markers in animal radiology images, hinder the effective use of image-based diagnostics by veterinarians.
Innovation Solution
The use of machine learning models, such as RapidReadNet and AdjustNet, to automate the processing and interpretation of pet radiology images, including the classification of abnormalities and correction of anatomical orientation without relying on DICOM metadata or laterality markers, utilizing techniques like convolutional neural networks and natural language processing to generate accurate and efficient diagnostic results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If veterinary radiologists manually review medical images, then diagnostic accuracy is maintained, but the process becomes time-consuming and inaccessible to most veterinarians
Solution Approach 1:
The patent replaces the mechanical system of manual image review by veterinary radiologists with an automated computer-based system using machine learning models. The system automatically processes radiographic images, performs classification, and generates diagnostic assessments without requiring manual intervention, thereby maintaining diagnostic accuracy while eliminating time consumption.
Solution Approach 2:
The patent enables veterinarians to perform diagnostic assessments independently through an automated system that processes images and provides interpretations without requiring consultation with specialized radiologists. The system serves itself by automatically correcting orientation issues, classifying abnormalities, and generating reports, making diagnostic capabilities accessible to non-specialist veterinarians.
2Adaptability or versatility
If veterinarians use image-based diagnostic techniques, then diagnostic capability is improved, but the limited number of trained radiologists prevents widespread adoption
Solution Approach 1:
The patent introduces an automated image analysis system as an intermediary between veterinarians and diagnostic interpretation. This intermediary handles the complex tasks of image processing, orientation correction, and abnormality classification, allowing veterinarians without specialized radiology training to effectively use image-based diagnostics. The system acts as a bridge that eliminates the need for veterinarians to undergo extensive specialized training.
3Productivity
If automated machine learning models are used to process images, then productivity increases, but the system must handle incorrect orientations and missing markers without specialized training data
Solution Approach 1:
The patent applies preliminary action by automatically correcting image orientation and identifying laterality markers before the main classification task. The system performs orientation correction and marker detection as preprocessing steps, ensuring that subsequent analysis operates on properly oriented images regardless of their initial state. This preliminary correction enables the system to handle incorrectly oriented images and those with missing markers effectively.
Data Source
AI summary
In one embodiment, the disclosure provides a computer-implemented method comprising: receiving a first labeled training data set comprising a first plurality of images each associated with a set of labels; programmatically training a machine learning neural Teacher model on the first labeled training data set; programmatically applying a machine learning model trained for NLP to an unlabeled data set comprising digital electronic representations of natural language text summaries of a second plurality of images, thereby generating a second labeled training data set comprising the second plurality of images; using the machine learning neural Teacher model, programmatically generating soft pseudo labels; programmatically generating derived labels using the soft pseudo labels; training one or more programmed machine learning neural Student models using the derived labels; receiving a target image; and applying an ensemble of one or more of the Student models to output one or more classifications of the target image.