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-12217425-B2
Publication Date
2025-02-04
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 provides an AI model for hierarchical lesion interpretation of medical images. The AI model first classifies a detected lesion into a suspicious class or a non-suspicious class, and when the lesion is classified as suspicious, the system further classifies it into a malignant class or a benign-hard class.
For lesions determined to be suspicious, the system outputs conditional-probability metrics and an interpretation result that reflects the hierarchical classification. The conditional-probability metrics include probabilities of being suspicious, being malignant, and being benign-hard, and the system selectively reflects lower-branch results in the final interpretation to avoid conflicting outputs.
The hierarchical classification includes independently trained branch classification models with hierarchically grouped labels. One grouping uses normal/benign-easy as non-suspicious, and another grouping uses benign-hard/cancer as suspicious. Feature extraction via a convolutional neural network style or ResNet-style feature extractor supports the branch classifications, and the interpretation outputs can include confidence calculations derived from conditional probabilities.
Claims Coverage
The document provides three independent claim types: computing device, method, and computer-readable storage medium. Across these, the core claim set includes one primary inventive concept: an AI model that classifies a detected lesion into predefined lesion types and outputs an interpretation result that includes both the classification class and a corresponding classification probability.
AI lesion classification with predefined lesion types and probability
The processor receives a target image, uses an AI model trained to classify a lesion detected in a medical image into one of predefined types of lesions, obtains a classification class of a target lesion among the predefined types of lesions, and outputs an interpretation result including the classification class and classification probability indicating a probability that the target lesion is classified into the classification class by the AI model.
Normal class outputs no lesion detected
The computing device outputs an interpretation result stating that no lesion is detected in the target image when the target image is classified as a normal class.
Lesion location in interpretation result
The computing device further provides, as part of the interpretation result, a location of a detected target lesion in a target image.
Training data with annotated classification labels
The computing device trains an AI model using training data that pairs each medical image with an annotated classification label indicating at least one lesion class present in the image.
Predefined lesion class set including non-suspicious, suspicious, normal, benign, malignant, benign-hard, or benign-easy
The classification classes include at least one specified class type such as non-suspicious, suspicious, normal, benign, malignant, benign-hard, or benign-easy.
Benign-hard case with conditional probability outputs
The computing device outputs an interpretation result indicating whether a target lesion is a benign-hard lesion and provides probabilities of being suspicious and of being benign-hard when the lesion is classified as a suspicious class and benign classes.
Overall, the claims cover an AI-based lesion interpretation workflow in which a trained AI model assigns a predefined lesion classification class to a detected lesion and outputs the interpretation result including that class and a classification probability. The dependent coverage further specifies normal/no-lesion handling, lesion location output, training label structure, example sets of lesion classes, and class-conditional probability output for a benign-hard scenario.
Stated Advantages
Improved reliability
Explainable outputs
Reduced unnecessary additional examinations
Documented Applications
Medical image lesion interpretation for modalities including X-ray, MRI, ultrasound, CT, mammography, and DBT.
Interested in licensing this patent?