Assessment of image quality for a medical diagnostics device
A medical diagnostic system can assess quality of a representation of a body part determined based on a response of the body part to exposure to electromagnetic waves, process the representation with a disease detection machine learning model to determine a certainty measure for a presence of a disease, determine a quality score for the representation based on the quality of the representation and the certainty measure, and discard the at least one representation based on the quality score. Combining machine learning in conjunction with one another, such as, for quality assessment and disease detection, can provide for more accurate image quality analysis, lead to faster medical imaging, and reduce the need to retake images or entirely re-perform medical imaging. The system can be easier to use, be more robust and faster than other systems by reducing the need to retake images while maintaining performance of the system.
1 . A retina camera comprising:
a housing;
a light source supported by the housing, the light source configured to irradiate an eye of a patient with light;
imaging optics supported by the housing, the imaging optics configured to receive light reflected by the eye;
an image detector array supported by the housing, the image detector array configured to receive light from the imaging optics and to sense the received light; and
an electronic processing circuitry supported by the housing, the electronic processing circuitry configured to:
generate at least one image of the eye based on signals received from the image detector array;
assess an image quality of the at least one image;
process the at least one image with a machine learning model to determine a certainty measure indicative of a presence of at least one disease from a plurality of diseases that the machine learning model has been trained to identify;
determine a quality score for the at least one image based on the image quality of the at least one image and the certainty measure indicative of the presence of the at least one disease, wherein determining the quality score comprises:
in response to the image quality not satisfying an image quality threshold indicative of sufficient image quality for disease detection, setting the quality score to a value that does not satisfy a quality threshold irrespective of the certainty measure satisfying a first certainty threshold indicative of sufficient likelihood that the at least one disease is present, and
in response to the certainty measure satisfying a second certainty threshold indicative of higher likelihood that the at least one disease is present than the first certainty threshold, setting the quality score to a value that satisfies the quality threshold irrespective of the image quality not satisfying the image quality threshold indicative of sufficient image quality for disease detection;
in response to a determination that the quality score does not satisfy the quality threshold, discard the at least one image; and
in response to a determination that the quality score satisfies the quality threshold, provide an indication of the presence of the at least one disease.
2 . The retina camera of claim 1 , wherein the electronic processing circuitry is configured to assess the image quality of the at least one image with another machine learning model different from the machine learning model.
3 . The retina camera of claim 1 , further comprising a display at least partially supported by the housing, wherein the electronic processing circuitry is further configured to provide the indication of the presence of the at least one disease on the display.
4 . The retina camera of claim 1 , wherein discarding the at least one image causes the at least one image to be retaken.
5 . The retina camera of claim 1 , wherein the image quality is assessed based on at least one of a probability or a confidence score.
6 . The retina camera of claim 1 , wherein the certainty measure is based on at least one of a probability generated by the machine learning model or a confidence score generated by the machine learning model.
7 . The retina camera of claim 1 , wherein the at least one image comprises a plurality of images determined from a video of the eye, and wherein discarding the at least one image causes use of another image from the plurality of images for detecting the presence of the at least one disease.
8 . The retina camera of claim 1 , wherein the electronic processing circuitry is configured to determine the quality score for the at least one image by:
in response to the certainty measure for the presence of the at least one disease not satisfying a third certainty threshold indicative of insufficient likelihood that the at least one disease is present, setting the quality score to a value that does not satisfy the quality threshold irrespective of the image quality satisfying the image quality threshold.
9 . The retina camera of claim 1 , wherein the electronic processing circuitry is configured to:
determine a confidence of the presence of the at least one disease based on a confidence score generated by the machine learning model and a confidence score generated by another machine learning model configured to assess the image quality of the at least one image; and
output the confidence of the presence of the at least one disease along with the indication of the presence of the at least one disease.
10 . A medical diagnostics system comprising:
an energy source configured to direct electromagnetic waves toward a body part of a patient;
a detector configured to sense a response of the body part to the electromagnetic waves; and
an electronic processing circuitry configured to:
generate at least one representation of the body part based on the response sensed by the detector;
assess a quality of the at least one representation;
process the at least one representation with a machine learning model to determine a certainty measure indicative of a presence of at least one disease from a plurality of diseases that the machine learning model has been trained to identify;
determine a quality score for the at least one representation based on the quality of the at least one representation and the certainty measure indicative of the presence of the at least one disease, wherein determining the quality score comprises:
in response to the quality of the at least one representation not satisfying a representation quality threshold indicative of sufficient quality for disease detection, setting the quality score to a value that does not satisfy a quality threshold irrespective of the certainty measure satisfying a first certainty threshold indicative of sufficient likelihood that the at least one disease is present, and
in response to the certainty measure satisfying a second certainty threshold indicative of higher likelihood that the at least one disease is present than the first certainty threshold, setting the quality score to a value that satisfies the quality threshold irrespective of the quality of the at least one representation not satisfying the representation quality threshold indicative of sufficient quality for disease detection;
in response to a determination that the quality score does not satisfy the quality threshold, discard the at least one representation; and
in response to a determination that the quality score satisfies the quality threshold, provide an indication of the presence of the at least one disease.
11 . The medical diagnostics system of claim 10 , wherein the energy source comprises a light source, an x-ray source, or a magnetic source.
12 . The medical diagnostics system of claim 10 , wherein the energy source comprises a light source, the detector comprises an image detector array, the body part comprises an eye, and the at least one representation comprises at least one image of the eye.
13 . The medical diagnostics system of claim 12 , wherein the at least one image comprises a plurality of images determined from a video of the eye, and wherein discarding the at least one representation causes using of another image from the plurality of images for detecting the presence of the at least one disease.
14 . The medical diagnostics system of claim 10 , further comprising a display, wherein the electronic processing circuitry is further configured to provide the indication of the presence of the at least one disease on the display.
15 . The medical diagnostics system of claim 10 , wherein the electronic processing circuitry is configured to assess the quality of the at least one representation with another machine learning model different from the machine learning model.
16 . The medical diagnostics system of claim 10 , wherein discarding the at least one representation causes the at least one representation to be retaken.
17 . The medical diagnostics system of claim 10 , wherein the quality of the at least one representation is assessed based on at least one of a probability or a confidence score.
18 . The medical diagnostics system of claim 10 , wherein the certainty measure is based on at least one of a probability generated by the machine learning model or a confidence score generated by the machine learning model.
19 . The medical diagnostics system of claim 10 , wherein the electronic processing circuitry is configured to determine the quality score for the at least one representation by:
in response to the certainty measure for the presence of the at least one disease not satisfying a third certainty threshold indicative of insufficient likelihood that the at least one disease is present, setting the quality score to a value that does not satisfy the quality threshold irrespective of the quality of the at least one representation satisfying the quality threshold.
20 . The medical diagnostics system of claim 10 , wherein the electronic processing circuitry is configured to:
determine a confidence of the presence of the at least one disease based on a confidence score generated by the machine learning model and a confidence score generated by another machine learning model configured to assess the quality of the at least one representation; and
output the confidence of the presence of the at least one disease along with the indication of the presence of the at least one disease.