SYSTEMS AND METHODS FOR PROCESSING ELECTRONIC IMAGES TO DETECT CONTAMINATION IN SPECIMEN PREPARATIONS
Systems and methods are disclosed for receiving one or more digital images associated with a tissue specimen, detecting one or more image regions from a background of the one or more digital images, determining a prediction, using a machine learning system, of whether at least one first image region of the one or more image regions comprises at least one external contaminant, the machine learning system having been trained using a plurality of training images to predict a presence of external contaminants and/or a location of any external contaminants present in the tissue specimen, and determining, based on the prediction of whether a first image region comprises an external contaminant, whether to process the image region using an processing algorithm.
1 . A computer-implemented method for training a machine learning system for processing digital images of a tissue specimen, the method comprising:
creating at least one synthetic foreign contaminant image;
generating a training dataset by combining a foreign contaminant dataset containing the at least one synthetic foreign contaminant image with at least one unaltered image having at least one foreign contaminant that is not synthetic;
detecting one or more potential foreign contaminant regions for each image of the training dataset;
extracting one or more features from at least one of the detected one or more potential foreign contaminant regions;
providing the one or more features from the at least one of the detected one or more potential foreign contaminant regions to the machine learning system for training the machine learning system, and determining, by the machine learning system, whether a foreign contaminant is present in the at least one of the detected one or more potential foreign contaminant regions.
2 . The computer-implemented method of claim 1 , wherein the at least one synthetic foreign contaminant image is created by mixing or composing at least one image region from a plurality of patients.
3 . The computer-implemented method of claim 2 , wherein the at least one image regions from the plurality of patients comprises a first image region having a first tissue type and a second image region having a second tissue type different from the first tissue type.
4 . The computer-implemented method of claim 2 , wherein the mixing or composing of the at least one image region from the plurality of patients is performed using generative adversarial networks.
5 . The computer-implemented method of claim 1 , wherein the detecting of the one or more potential foreign contaminant regions for each image of the training dataset comprises thresholding based on color intensity, texture features, and/or Otsu's method.
6 . The computer-implemented method of claim 1 , wherein the detecting of the one or more potential foreign contaminant regions for each image of the training dataset comprises using one or more segmentation algorithms.
7 . The computer-implemented method of claim 1 , wherein the detecting of the one or more potential foreign contaminant regions for each image of the training dataset comprises using bounding box detection.
8 . The computer-implemented method of claim 1 , wherein the one or more features extracted from the at least one of the detected one or more potential foreign contaminant regions comprise one or more of convolutional neural network (CNN) features, SIFT features, or SURF features.
9 . The computer-implemented method of claim 1 , wherein the machine learning system is implemented by a convolutional neural network (CNN), Region CNN (R-CNN), Faster R-CNN, Mask R-CNN, Recurrent Neural Network (RNN), Multi-Layer Perceptron (MLP), Convolutional Graph Neural Network, and/or Relationship Neural Network.
10 . A system for training a machine learning algorithm for processing digital images of a tissue specimen, the system comprising:
at least one memory storing instructions; and
at least one processor configured to execute the instructions to perform operations comprising:
creating at least one synthetic foreign contaminant image;
generating a training dataset by combining a foreign contaminant dataset containing the at least one synthetic foreign contaminant image with at least one unaltered image having at least one foreign contaminant that is not synthetic;
detecting one or more potential foreign contaminant regions for each image of the training dataset;
extracting one or more features from at least one of the detected one or more potential foreign contaminant regions;
providing the one or more features from the at least one of the detected one or more potential foreign contaminant regions to the machine learning algorithm for training the machine learning algorithm, and determining, by the machine learning algorithm, whether a foreign contaminant is present in the at least one of the detected one or more potential foreign contaminant regions.
11 . The system of claim 10 , wherein the at least one synthetic foreign contaminant image is created by mixing or composing at least one image region from a plurality of patients.
12 . The system of claim 11 , wherein the at least one image regions from the plurality of patients comprises a first image region having a first tissue type and a second image region having a second tissue type different from the first tissue type.
13 . The system of claim 11 , wherein the mixing or composing of the at least one image region from the plurality of patients is performed using generative adversarial networks.
14 . The system of claim 10 , wherein the detecting of the one or more potential foreign contaminant regions for each image of the training dataset comprises thresholding based on color intensity, texture features, and/or Otsu's method.
15 . The system of claim 10 , wherein the detecting of the one or more potential foreign contaminant regions for each image of the training dataset comprises using one or more segmentation algorithms or bounding box detection.
16 . The system of claim 10 , wherein the one or more features extracted from the at least one of the detected one or more potential foreign contaminant regions comprise one or more of convolutional neural network (CNN) features, SIFT features, or SURF features.
17 . The system of claim 10 , wherein the machine learning algorithm is implemented by a convolutional neural network (CNN), Region CNN (R-CNN), Faster R-CNN, Mask R-CNN, Recurrent Neural Network (RNN), Multi-Layer Perceptron (MLP), Convolutional Graph Neural Network, and/or Relationship Neural Network.
18 . At least one non-transitory computer readable medium for processing an electronic image corresponding to a tissue specimen, the at least one non-transitory computer readable medium storing instructions which, when executed by one or more processors, cause the one or more processors to perform operations comprising:
creating at least one synthetic foreign contaminant image;
generating a training dataset by combining a foreign contaminant dataset containing the at least one synthetic foreign contaminant image with at least one unaltered image having at least one foreign contaminant that is not synthetic;
detecting one or more potential foreign contaminant regions for each image of the training dataset;
extracting one or more features from at least one of the detected one or more potential foreign contaminant regions;
providing the one or more features from the at least one of the detected one or more potential foreign contaminant regions to a machine learning algorithm for training the machine learning algorithm, and determining, by the machine learning algorithm, whether a foreign contaminant is present in the at least one of the detected one or more potential foreign contaminant regions.
19 . The at least one non-transitory computer readable medium of claim 18 , wherein the at least one synthetic foreign contaminant image is created by mixing or composing at least one image region from a plurality of patients.
20 . The at least one non-transitory computer readable medium of claim 19 , wherein the at least one image regions from the plurality of patients comprises a first image region having a first tissue type and a second image region having a second tissue type different from the first tissue type.