Method for discriminating suspicious lesion in medical image, method for interpreting medical image, and computing device implementing the methods
Inventors
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-11935237-B2
Publication Date
2024-03-19
Expiration Date
Abstract
A method for interpreting an input image by a computing device operated by at least one processor is provided. The method for interpreting an input image comprises storing an artificial intelligent (AI) model that is trained to classify a lesion detected in the input image as suspicious or non-suspicious and, under a condition of being suspicious, to classify the lesion detected in the input image as malignant or benign-hard representing that the lesion is suspicious but determined to be benign, receiving an analysis target image, by using the AI model, obtaining a classification class of a target lesion detected in the analysis target image and, when the classification class is the suspicious, obtaining at least one of a probability of being suspicious, a probability of being benign-hard, and a probability of malignant for the target lesion, and outputting an interpretation result including at least one probability obtained for the target lesion.
Core Innovation
The invention relates to an artificial intelligent framework for medical image interpretation of a target lesion detected in a target image. The framework outputs an interpretation result including at least one classification class of the target lesion and at least one classification probability indicating a probability that the target lesion is classified into the at least one classification class by the AI model.
The disclosed approach uses hierarchical classification to represent suspiciousness before malignancy assessment, producing interpretation probabilities such as a probability of being suspicious and conditional probabilities associated with malignant or benign-hard classes when a lesion is classified as suspicious. In the disclosed hierarchical scheme, a first branch classification model classifies the lesion as suspicious or non-suspicious, and when the lesion is classified as suspicious, a second branch classification model further classifies the lesion as malignant or benign-hard.
The framework further expands the classification scheme by adding additional branches to classify non-suspicious into normal or benign-easy. The described framework uses independently trained branch classifiers that share a feature extractor, so the branch classifiers operate on common features while producing different classification outcomes for different levels of the hierarchical classification.
Claims Coverage
The independent claims are directed to a computing-device method for interpreting a target image with an AI model that outputs classification classes and classification probabilities for a detected target lesion, and to a computing device configured to perform the same. The inventive features include hierarchical lesion classification with specific lesion class groupings and the output of interpretation results including classification probabilities, with dependent claim refinements adding additional classification classes and conditional interpretation outputs.
Hierarchical lesion classification using branch classification models
Using an AI model including a first branch classification model to classify a target lesion as suspicious or non-suspicious and, when classified as suspicious, a second branch classification model to classify the lesion as malignant or benign-hard.
Shared feature extractor with independently trained branch classifiers
Employing independently trained branch classification models that share a feature extractor for obtaining classification classes for the detected target lesion.
Interpretation result with classification probabilities
Outputting an interpretation result including the at least one classification class of the target lesion and at least one classification probability indicating a probability that the target lesion is classified into one or each of the at least one classification class.
Additional classification branches for normal and benign-easy
Classifying non-suspicious lesions into a normal class or a benign-easy class using additional branches.
Conditional malignancy interpretation output and lesion location
Providing an interpretation result including a location of a target lesion and a probability that it is malignant when the target lesion is classified in a malignant class.
Normal-class output indicating no lesion detected
Outputting an interpretation result indicating that no lesion is detected in a target image when the target image is classified as a normal class.
Enumeration of selectable classification classes including benign-easy
Allowing selection of the at least one classification class from non-suspicious, suspicious, normal, benign, malignant, benign-hard, and benign-easy classes.
Overall, the claim coverage focuses on AI-based interpretation that outputs lesion classification classes together with classification probabilities, implemented via a hierarchical suspiciousness-first branch structure (suspicious vs non-suspicious, then malignant vs benign-hard) with shared features across independently trained branch classification models. Dependent claim refinements expand the output to include normal and benign-easy categories, conditional malignant interpretation including lesion location and malignancy probability, explicit no lesion detected output for normal classification, and an enumerated set of classification classes including benign-easy.
Stated Advantages
Reduces false positives.
Avoids missing benign-but-suspicious lesions.
Reduces unnecessary follow-up examinations.
Documented Applications
Medical image interpretation framework for lesions in medical image modalities, including mammography, ultrasound, MRI, CT, X-ray, and DBT.
Interested in licensing this patent?