Correlating regions of interest
Methods and systems for identifying a region of interest in breast tissue use artificial intelligence to confirm that a target lesion identified during initial imaging the breast tissue has been identified in a subsequent imaging session. A computing system operating a lesion matching engine uses a machine learning classifier algorithm trained on cases of initial and subsequent two-dimensional or three-dimensional images of lesions. The lesion matching engine analyzes a target lesion identified with initial and a potential lesion identified with current imaging to determine a likelihood that the target lesion is the same as the potential lesion. A confidence level indicator for the lesion match is presented on a display of a computing device to aid a healthcare provider in locating a lesion in breast tissue.
1 . A method of locating of a lesion within a breast, the method comprising:
receiving, at a computing device, an initial image of a target lesion within the breast, wherein the initial image was generated during an initial imaging session;
receiving, at the computing device, a current image of the breast obtained during a current biopsy session;
receiving, at the computing device, an indication of a location of a potential lesion in the current image;
analyzing, with a machine learning lesion classifier, the potential lesion to compare the potential lesion to the target lesion and determine a level of confidence that the potential lesion corresponds to the target lesion, wherein the machine learning lesion classifier analyzes lesions for correlations in one or more of density, stiffness, shape, size, and texture;
outputting an indicator of the level of confidence for display on a graphical user interface of a biopsy computing device; and
based at least partially on the level of confidence, navigating a biopsy device, via the biopsy computing device, to the potential lesion during the current biopsy session.
2 . The method of claim 1 , further comprising:
performing a biopsy on the potential lesion.
3 . The method of claim 1 , further comprising:
recording location coordinates of the target lesion; and
recording location coordinates of the potential lesion,
wherein analyzing the potential lesion comprises comparing location coordinates of the target lesion and location coordinates of the potential lesion.
4 . The method of claim 3 , wherein the location coordinates of the potential lesion comprise a clock position relative to a nipple of the breast, a depth from a surface of the breast, and a distance from the nipple.
5 . The method of claim 1 , wherein the initial image and the current image are two-dimensional images or are obtained using digital breast tomosynthesis.
6 . The method of claim 1 , further comprising providing the initial image and the current image as input to the machine learning lesion classifier.
7 . The method of claim 1 , wherein analyzing the potential lesion is performed using an artificial intelligence system trained using a library of digital breast tomosynthesis images.
8 . The method of claim 7 , wherein the artificial intelligence system is an unsupervised machine learning system.
9 . The method of claim 1 , wherein analyzing the potential lesion comprises comparing form factors of the target lesion and the potential lesion.
10 . The method of claim 1 , wherein the indicator comprises displaying at least one of a shape, a color, a numerical value, or a reference on the received image.
11 . A lesion identification system comprising:
a processing device; and
a memory storing instructions that, when executed by the processing device, facilitate performance of operations, comprising:
accessing an initial x-ray image of a breast, the initial x-ray image including an identified lesion indicated with a visual marker;
receiving a current x-ray image of the breast, the current x-ray image including an indication of a potential lesion, wherein the current x-ray image is obtained during a current biopsy session;
analyzing the potential lesion and the identified lesion using an artificial intelligence lesion classifier, wherein the artificial intelligence lesion classifier is trained on digital breast tomosynthesis images, and wherein the artificial intelligence lesion classifier analyzes lesions for correlations in one or more of density, stiffness, shape, margins, orientation, texture, pattern, size, and depth within the breast;
generating a confidence score indicating a likelihood that the potential lesion in the current x-ray image matches the identified lesion in the initial x-ray image;
displaying an output associated with the confidence score on a graphical user interface; and
based at least partially on the confidence score, navigating a biopsy device, via the graphical user interface, towards the potential lesion during the current biopsy session.
12 . The system of claim 11 , wherein the operations further comprise accessing a first set of coordinates identifying a location of the identified lesion while the breast is under compression; translating the first set of coordinates into a second set of coordinates identifying a predicted location of the identified lesion while the breast is not under compression; identifying a region of interest in the current x-ray image corresponding to the second set of coordinates; and identifying the potential lesion in the current x-ray image.
13 . The system of claim 11 , wherein the artificial intelligence lesion classifier is trained on two-dimensional images.
14 . The system of claim 11 ,
wherein the artificial intelligence lesion classifier uses an unsupervised machine learning system.
15 . The system of claim 11 , wherein the initial x-ray image was generated during a screening session.
16 . The system of claim 11 , wherein the current x-ray image is received from an x-ray system and the confidence score is communicated to a display in communication with the x-ray system.
17 . The system of claim 16 , wherein a visual indication of the confidence score is displayed on the display in communication with the x-ray system as a visual indicator on the current x-ray image.
18 . A non-transitory machine-readable storage medium, comprising executable instructions that, when executed by a processor, facilitate performance of operations, comprising:
obtaining data for a target lesion from a data store, wherein the data was obtained with x-ray imaging and includes at least an initial image of the target lesion and coordinates for a location of the target lesion within a breast;
recording a current image of the breast obtained by x-ray imaging during a current biopsy session;
identifying a general area of interest in the current image of the breast based on the coordinates of the target lesion;
identifying a potential lesion in the general area of interest;
analyzing, using an artificial intelligence lesion classifier, the potential lesion to compare the potential lesion to the target lesion and determine a level of confidence that the potential lesion corresponds to the target lesion, wherein the artificial intelligence lesion classifier analyzes lesions for correlations in one or more of density, stiffness, shape, size, and texture;
outputting an indicator of the level of confidence on a graphical user interface; and
based at least partially on the level of confidence, navigating a biopsy device, via the graphical user interface, towards the potential lesion during the current biopsy session.
19 . The non-transitory machine-readable storage medium of claim 18 , wherein the operations further comprise:
recording the initial image using digital breast tomosynthesis;
receiving an indication on the initial image of the target lesion within the breast;
determining coordinates for a location of the target lesion; and
saving the indication and coordinates of the target lesion in the data store.
20 . A lesion identification system comprising:
a processing device; and
a memory storing instructions that, when executed by the processing device, facilitate performance of operations, comprising:
accessing an initial x-ray image of a breast, the initial x-ray image including an identified lesion indicated with a visual marker;
accessing a first set of coordinates identifying a location of the identified lesion while the breast is under compression;
translating the first set of coordinates into a second set of coordinates and identifying a predicted location of the identified lesion while the breast is not under compression;
receiving a current x-ray image of the breast, the current x-ray image including an indication of a potential lesion;
identifying a region of interest in the current x-ray image corresponding to the second set of coordinates;
identifying the potential lesion in the current x-ray image;
analyzing the potential lesion and the identified lesion using an artificial intelligence lesion classifier;
generating a confidence score indicating a likelihood that the potential lesion in the current x-ray image matches the identified lesion in the initial x-ray image;
displaying an output associated with the confidence score on a graphical user interface; and
based at least partially on the confidence score, navigating a biopsy device, via the graphical user interface, towards the potential lesion.