System and method for classifying dermatological images using machine learning
Systems and methods using machine learning for classifying images as being sufficient for medical diagnosis. An example of the method includes: receiving a dataset comprising a plurality of medical images; receiving, from a first single source, a respective label for each one of the plurality of medical images, the respective label being a positive response versus a negative response; dividing each one of the plurality of medical images into a plurality of medical image segments; associating each one of the plurality of medical image segments with an image segment label based on the respective label for the respective medical image being divided; and training a machine learning model using: the plurality of medical images, the respective label for each one of the plurality of medical images, the plurality of medical image segments, and the respective image segment label of each one of the plurality of medical image segments.
1. A method for training a machine learning model, the method comprising:
receiving a dataset comprising a plurality of medical images;
receiving, from a first single source, a respective label for each one of the plurality of medical images, the respective label being a positive response versus a negative response, wherein the positive response as the respective label for the respective medical image is an indication that the medical image has a quality sufficient for medical diagnosis, without indicating the medical diagnosis on the medical image;
receiving a gaze profile for each one of the plurality of medical images from the first single source;
dividing each one of the plurality of medical images into a plurality of medical image segments;
associating each one of the plurality of medical image segments with an image segment label based on the respective label for the respective medical image being divided;
for each medical image of the plurality of medical image segments from the plurality of medical images, when the respective image segment label is a positive response: generating a gaze label based on the number of gaze hits for each one of the plurality of medical image segments for the medical image based on the received gaze profile for the medical image, and associating each one of the plurality of medical image segments with the respective gaze label; and
sending a first set of labelled datasets to the machine learning model to train the machine learning model to identify image segments as having the quality sufficient for medical diagnosis for the first single source, wherein each labelled dataset in the first set includes a respective medical image segment of a medical image associated with a gaze label and an image segment label, the gaze label indicates that the number of gaze hit of the medical image segment being above a threshold, and the image segment label is the positive response.
2. The method as claimed in claim 1 , wherein the gaze profile is generated based on tracking eye movements of the first single source and comprises a distribution of gaze hits from the first single source for the respective medical image.
3. The method as claimed in claim 2 , wherein the respective label for each one of the plurality of medical images comprises a binary value.
4. The method as claimed in claim 3 , wherein the binary value represents the positive response versus the negative response.
5. The method as claimed in claim 4 , wherein the respective image segment label for each one of the plurality of medical image segments comprises an image segment binary value that represents the positive response versus the negative response, wherein the image segment binary value is based on the binary value of the respective label for the medical image containing the medical image segment.
6. The method as claimed in claim 1 , wherein the machine learning model is not trained to make any medical diagnosis based on any of the plurality of medical images.
7. The method as claimed in claim 1 , wherein each one of the plurality of medical image segments has a minimum dimension of 224 pixels by 224 pixels.
8. The method as claimed in claim 7 , wherein each one of the plurality of medical image segments has a dimension of 256 pixels by 256 pixels.
9. The method of claimed in claim 1 , further comprising, before receiving the respective label for each one of the plurality of medical images, re-sizing each one of the plurality of medical images to a minimum of 720 pixels on one side of the respective medical image.
10. The method as claimed in claim 9 , wherein the re-sizing of each one of the plurality of medical images is to 1024 pixels on the one side of the respective medical image.
11. The method as claimed in claim 1 , wherein the machine learning model comprises at least one of: a support vector machine (SVM), linear regression, or a convolutional neural network (CNN).
12. The method as claimed in claim 1 , wherein the medical images include dermatological images.
13. The method as claimed in claim 1 , further comprising:
receiving, from a second single source, a respective second label for each one of the plurality of medical images, the respective second label being the positive response versus the negative response;
associating each one of the plurality of medical image segments of each one of the plurality of medical images with a second image segment label based on the respective second label for the respective medical image being divided; and
receiving a second gaze profile for each one of the plurality of medical images from the second single source;
generating a second gaze label based on the number of gaze hits for each one of the plurality of medical image segments for the medical image based on the received second gaze profile for the medical image; and
sending a second set of labelled datasets to the machine learning model to train the machine learning model using the respective second gaze label for each of the plurality of medical images and the second image label for each of the plurality of medical image segments from the plurality of medical images.
14. The method as claimed in claim 1 , further comprising:
receiving additional information regarding each one of the plurality of medical images from the first single source; and
associating each one of the plurality of medical images with the respective additional information;
wherein the training of the machine learning model further comprises training the machine learning model for the first single source using: the respective additional information of each one of the plurality of medical images.
15. The method as claimed in claim 14 , wherein the additional information regarding each one of the plurality of medical images comprises one or more of: a dosage information, or a treatment recommendation.
16. The method as claimed in claim 1 , wherein at least one of the plurality of medical images is captured from a camera.
17. The method as claimed in claim 1 , wherein the method is performed by a processing device.
18. The method as claimed in claim 1 , the method further comprising: sending a second set of labelled datasets to the machine learning model to train the machine learning model, wherein each labelled dataset in the second set includes a respective medical image segment of a medical image associated with a gaze label and an image segment label, the gaze label indicates that the number of gaze hit of the medical image segment being less than a threshold, and the image segment label is the negative response.
19. A system for training a machine learning model, the system comprising:
a processing device; and
a memory coupled to the processing device, the memory storing machine-executable instructions that, when executed by the processing device, cause the processing device to:
receive a dataset comprising a plurality of medical images;
receive, from a first single source, a respective label for each one of the plurality of medical images, the respective label being a positive response versus a negative response, wherein the positive response as the respective label for the respective medical image is an indication that the medical image has a quality sufficient for medical diagnosis, without indicating the medical diagnosis on the medical image;
receive a gaze profile for each one of the plurality of medical images from the first single source;
divide each one of the plurality of medical images into a plurality of medical image segments;
associate each one of the plurality of medical image segments with an image segment label based on the respective label for the respective medical image being divided; and
send a first set of labelled datasets to the machine learning model to train the machine learning model to identify image segments as having the quality sufficient for medical diagnosis for the first single source, wherein each labelled dataset in first set includes a respective medical image segment of a medical image associated with a gaze label and an image segment label, the gaze label indicates that the number of gaze hit of the medical image segment being above a threshold, and the image segment label is the positive response.
20. A non-transitory computer readable medium containing program instructions for causing a processing device to perform a method of training a machine learning model, the instructions including:
instructions for receiving a dataset comprising a plurality of medical images;
instructions for receiving, from a first single source, a respective label for each one of the plurality of medical images, the respective label being a positive response versus a negative response, wherein the positive response as the respective label for the respective medical image is an indication that the medical image has a quality sufficient for medical diagnosis, without indicating the medical diagnosis on the medical image;
instructions for receiving a gaze profile for each one of the plurality of medical images from the first single source;
instructions for dividing each one of the plurality of medical images into a plurality of medical image segments;
instructions for associating each one of the plurality of medical image segments with an image segment label based on the respective label for the respective medical image being divided;
for each medical image of the plurality of medical image segments from the plurality of medical images, when the respective image segment label is a positive response: instructions for generating a gaze label based on the number of gaze hits for each one of the plurality of medical image segments for the medical image based on the received gaze profile for the medical image, and instructions for associating each one of the plurality of medical image segments with the respective gaze label; and
instructions for sending a first set of labelled datasets to the machine learning model to train the machine learning model to identify image segments as having the quality sufficient for medical diagnosis for the first single source, wherein each labelled dataset in the first set includes a respective medical image segment of a medical image associated with a gaze label and an image segment label, the gaze label indicates that the number of gaze hit of the medical image segment being above a threshold, and the image segment label is the positive response.