Automatic detection of erosions and ulcers in Crohn's capsule endoscopy
Inventors
SOUSA FERREIRA, João Pedro • DA QUINTA E COSTA DE MASCARENHAS SARAIVA, Miguel José • CASAL CARDOSO, Hélder Manuel • GONCALVES DE MACEDO, Manuel Guilherme • LIMA AFONSO, João Pedro • RIBEIRO ANDRADE, Ana Patricia • NATAL JORGE, Renato Manue • LAGES PARENTE, Marco Paulo
Assignees
DIGESTAID Digestive Artificial Intelligence Development Ltda
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-12573031-B2
Publication Date
2026-03-10
Expiration Date
Abstract
The present invention relates to a computer-implemented method capable of automatically detecting small bowel and colonic ulcers and erosions in Crohn's capsule endoscopy image data, by classifying pixels as lesion or non-lesion, using a convolutional image feature extraction step followed by a classification step and indexing such lesions in the set of one or more classes.
Core Innovation
The invention is a computer-implemented method capable of automatically identifying erosions and ulcers in colon capsule colonoscopy medical images by classifying the pixels as erosions or ulcers. It uses a pixel classification neural network component arranged after a convolutional neural network image feature extraction component to obtain classification of erosion and ulcer pixels from colon capsule colonoscopy images.
The method selects subsets of colon capsule colonoscopy images such that each subset comprises only images from the same patient, and selects another subset as a validation set that does not overlap with previously selected subsets. It performs pre-training on each chosen subset using a plurality of combinations of convolutional neural network image feature extraction component and a subsequent classification neural network component for pixel classification, and stops early when the scores do not improve over a given number of epochs.
The method evaluates the performance of each combination, repeats the pre-training on new different subsets with another networks combination and training hyperparameters, and selects the architecture combination that performs best during pre-training. It then fully trains and validates the selected architecture combination using the entire set of colon capsule colonoscopy images to obtain an optimized architecture combination, predicts erosions and ulcers using the optimized architecture combination, and outputs the classification to an output collect module for validation by at least one of another neural network, any other computational system, or optionally a physician expert, with corrected predictions stored in a storage component.
Claims Coverage
The independent claim covers one inventive workflow with patient-disjoint subset selection, pre-training with convolutional feature extraction plus pixel classification, early stopping, architecture selection based on performance evaluation, full training on all images, pixel-level prediction, and third-party validation with storage of corrected predictions.
Patient-disjoint subset selection for training and validation
Selecting subsets of all colon capsule colonoscopy images, each subset comprising only images from the same patient, and selecting another subset as validation set that does not overlap with previously selected subsets.
Pre-training with CNN feature extraction and pixel classification
Pre-training each chosen subset with a plurality of combinations of convolutional neural network image feature extraction component and a subsequent classification neural network component for pixel classification as erosions and ulcers.
Early stopping when scores do not improve
Stopping early when the scores do not improve over a given number of epochs.
Performance evaluation and repetition with new subsets and hyperparameters
Evaluating the performance of each combination and repeating the pre-training on new different subsets with another networks combination and training hyperparameters.
Selecting the best architecture combination for full training
Selecting the architecture combination that performs best during pre-training, then fully training and validating the selected architecture combination using the entire set of colon capsule colonoscopy images to obtain an optimized architecture combination.
Prediction and output collection with third-party validation
Predicting erosions and ulcers using the optimized architecture combination and outputting the classification to an output collect module for validation by another neural network, any other computational system, or optionally a physician expert.
Storing corrected predictions
Storing the corrected prediction into a storage component.
The claim covers automatic pixel classification of erosion and ulcer findings in colon capsule colonoscopy images using patient-disjoint training and validation subsets, pre-training across combinations of CNN feature extraction and pixel classification, selection of a best-performing architecture for full training, and third-party validation that corrects wrong predictions and stores the corrected predictions.
Stated Advantages
Enables automatic identifying of erosions and ulcers in colon capsule colonoscopy medical images by classifying pixels as erosions or ulcers.
Uses an optimized architecture combination obtained by pre-training evaluation and full training and validation using the entire set of images.
Supports validation of classification outputs by another neural network, another computational system, and/or optionally a physician expert, with corrected predictions stored.
Documented Applications
Automatically identifying erosions and ulcers in colon capsule colonoscopy medical images using pixel classification for colon capsule colonoscopy.
Validating and correcting classification outputs for erosion and ulcer detection using a third-party validation step and storing corrected predictions.
Interested in licensing this patent?