Medical object detection and identification via machine learning
AssigneeInternational Business Machines Corporation
InventorsYoel Shoshan, Vadim Ratner
Life of this patent
Estimating the term…
Abstract
An approach for improving object detection performance by using bilateral organ information disclosed. The approach comprises introducing a penalty term to an object localization model. The penalty term encourages the model to identify same physical regions (actual same physical location in the patient's body) in multiple images, which create an image of the same organ. The approach includes a sub-component of the model that can output an entity embedding, in additional to the conventional classification and localization prediction. During optimization process, similar entity embedding for the same logical entity are encouraged, and similar entity embedding for different logical entities are discouraged.
Claims20
A computer-implemented method for optimizing a loss function associated with image analysis, the computer-method comprising:
- identifying one or more ground truth entities based on one or more image data;
- representing, using a machine-learning model, one or more anchors associated with the ground truth entities;
- creating, on top of a classification of the machine learning model, a first identity vector and a second identity vector based on one or more modified anchors;
- determining whether the one or more ground truth entities match each other based on a prediction by a neural network;
- in responsive to the one or more ground truth entities do match each other, calculating a penalty value based on the first identity vector and the second identity vector;
- optimizing a loss function of the machine learning model based on the penalty value, wherein the penalty value minimizes the loss function;
- and
- in responsive to the one or more ground truth entities do not match each other, penalizing the loss function based on an inverse length of a predefined intersection distance metric.
A computer program product for optimizing a loss function associated with image analysis, the computer program product comprising:
- one or more computer readable and non-transitory storage media and program instructions stored on the one or more computer readable and non-transitory storage media, the program instructions comprising:
- program instructions to identify one or more ground truth entities based on one or more image data;
- program instructions to represent, using a machine-learning model, one or more anchors associated with the ground truth entities;
- program instructions to create, on top of a classification of the machine learning model, a first identity vector and a second identity vector based on one or more modified anchors;
- program instructions to determine whether the one or more ground truth entities match each other based on a prediction by a neural network;
- in responsive to the one or more ground truth entities do match each other, program instructions to calculate a penalty value based on the first identity vector and the second identity vector;
- program instructions to optimize a loss function of the machine learning model based on the penalty value wherein the penalty value minimizes the loss function;
- and
- in responsive to the one or more ground truth entities do not match each other, program instructions to penalize the loss function based on an inverse length of a predefined intersection distance metric.
A computer system for optimizing a loss function associated with image analysis, the computer system comprising:
- one or more computer processors;
- one or more computer readable storage media;
- program instructions stored on the one or more computer readable storage media for execution by at least one of the one or more computer processors, the program instructions comprising:
- program instructions to identify one or more ground truth entities based on one or more image data;
- program instructions to represent, using a machine-learning model, one or more anchors associated with the ground truth entities;
- program instructions to create, on top of a classification of the machine learning model, a first identity vector and a second identity vector based on one or more modified anchors;
- program instructions to determine whether the one or more ground truth entities match each other based on a prediction by a neural network;
- in responsive to the one or more ground truth entities do match each other, program instructions to calculate a penalty value based on the first identity vector and the second identity vector;
- program instructions to optimize a loss function of the machine learning model based on the penalty value wherein the penalty value minimizes the loss function;
- and
- in responsive to the one or more ground truth entities do not match each other, program instructions to penalize the loss function based on an inverse length of a predefined intersection distance metric.
Dependencies are inferred from the claim text.