Hypothesizing and verification networks and methods for specimen classification
A method of characterizing a specimen as containing hemolysis, icterus, or lipemia is provided. The method includes capturing one or more images of the specimen, wherein the one or more images include a serum or plasma portion of the specimen. Pixel data is generated by capturing the image. The pixel data of the one or more images of the specimen is processed using a first network executing on a computer to predict a classification of the serum or plasma portion, wherein the classification comprises hemolysis, icterus, and lipemia. The predicted classification is verified using one or more verification networks. Quality check modules and specimen testing apparatus adapted to carry out the method are described, as are other aspects.
1 . A method of characterizing a specimen, comprising:
receiving an image of the specimen, the image including pixel data representative of the specimen;
providing a multi-stage hierarchical network system having:
a first, hypothesizing network trained to classify multiple categories including hemolysis, icterus, and lipemia (HIL) and multiple classes of categories including at least multiple classes of hemolysis; and
a plurality of HIL verification networks, each verification network trained solely on a respective one of the multiple categories including a respective verification network trained solely on each one of the multiple hemolysis classes, at least one verification network comprising a convolutional neural network;
processing pixel data of the image of the specimen using the first hypothesizing network executing on a computer to predict a classification of the specimen, wherein the classification comprises at least a category of hemolysis, icterus, or lipemia;
verifying the predicted classification of the first hypothesizing network using one of the verification networks trained solely on the predicted classification by processing pixel data of the image of the specimen using the one of the verification networks; and
reporting a failure case if the one of the verification networks used to verify the predicted classification of the first hypothesizing network fails to verify the predicted classification.
2 . The method of claim 1 , further comprising:
verifying the predicted classification by a verification network trained solely on hemolysis in response to the first hypothesizing network predicting a hemolysis classification of the specimen;
verifying the predicted classification by a verification network trained solely on icterus in response to the first hypothesizing network predicting an icterus classification of the specimen; and
verifying the predicted classification by a verification network trained solely on lipemia in response to the first hypothesizing network predicting a lipemia classification of the specimen.
3 . The method of claim 1 , wherein:
processing pixel data further comprises predicting a classification index for one of hemolysis, icterus, and lipemia; and
verifying the predicted classification further comprises verifying the predicted classification index using one of the plurality of HIL verification networks trained solely on the predicted classification index.
4 . The method of claim 1 , wherein:
processing pixel data further comprises predicting a hemolysis classification index; and
verifying the predicted classification further comprises verifying the predicted hemolysis classification index using an index verification network of the plurality of HIL verification networks trained solely on the predicted hemolysis classification index.
5 . The method of claim 1 , wherein:
processing pixel data further comprises predicting an icterus classification index; and
verifying the predicted classification further comprises verifying the predicted icterus classification index using an index verification network of the plurality of HIL verification networks trained solely on the predicted icterus classification index.
6 . The method of claim 1 , wherein:
processing pixel data further comprises predicting a lipemia classification index; and
verifying the predicted classification further comprises verifying the predicted lipemia classification index using an index verification network of the plurality of HIL verification networks trained solely on the predicted lipemia classification index.
7 . The method of claim 1 , wherein the classification further comprises a normal classification.
8 . The method of claim 7 , further comprising verifying the normal classification using a verification network trained on a normal classification.
9 . The method of claim 1 , wherein processing pixel data of the image of the specimen using the first hypothesizing network comprises processing pixel data of one or more images of the specimen using a segmentation network.
10 . The method of claim 1 , wherein processing pixel data of the image of the specimen using the first hypothesizing network comprises processing pixel data of one or more images of the specimen using a classification network.
11 . The method of claim 1 , wherein processing pixel data comprises:
identifying a serum or plasma portion in the specimen using a deep semantic segmentation network; and
predicting a classification index of the serum or plasma portion of the specimen using the deep semantic segmentation network based at least partially on color represented by the pixel data of the serum or plasma portion.
12 . The method of claim 1 , wherein the first hypothesizing network comprises an architecture including at least eleven dense block layers.
13 . The method of claim 1 , further comprising generating a confidence level in response to verifying the predicted classification using one or more of the plurality of HIL verification networks.
14 . The method of claim 1 , further comprising generating a confidence level in response to verifying the predicted classification using one or more of the plurality of HIL verification networks and generating a signal in response to the confidence level being below a predetermined level.
15 . The method of claim 1 , wherein the plurality of verification networks comprises two or more convolutional neural networks.
16 . A non-transient computer-readable data storage medium comprising executable program code that includes a system, comprising:
a multi-stage hierarchical network system having:
a first, hypothesizing network trained to classify multiple categories including hemolysis, icterus, and lipemia (HIL) and multiple classes of categories including at least multiple classes of hemolysis; and
a plurality of HIL verification networks, each verification network trained solely on a respective one of the multiple categories including a respective verification network trained solely on each one of the multiple hemolysis classes, the at least one verification network comprising a convolutional neural network;
wherein the multi-stage hierarchical network system is configured to:
receive an image of the specimen, the image including pixel data representative of the specimen;
process pixel data of the image of the specimen using the first hypothesizing network executing on a computer to predict a classification of the specimen, wherein the classification comprises at least one of hemolysis, icterus, or lipemia;
verify the predicted classification of the first hypothesizing network using one of the verification networks trained solely on the predicted classification by processing pixel data of the image of the specimen using the one of the verification networks; and
report a failure case if the verification network used to verify the predicted classification of the first hypothesizing network fails to verify the predicted classification.
17 . The non-transient computer-readable data storage medium of claim 16 , wherein the multi-stage hierarchical network system is further configured and operative to process pixel data to predict a normal classification of the serum or plasma portion.
18 . The non-transient computer-readable data storage medium of claim 16 , wherein the multi-stage hierarchical network system is configured and operative to:
process pixel data to predict a classification index; and
verify the predicted classification index using one of the plurality of HIL verification networks trained solely on the predicted classification index.
19 . The non-transient computer-readable data storage medium of claim 16 , wherein the multi-stage hierarchical network system is configured and operative to:
verify the predicted classification by a verification network trained solely on hemolysis in response to the first hypothesizing network predicting a hemolysis classification for the specimen;
verify the predicted classification by a verification network trained solely on icterus in response to the first hypothesizing network predicting an icterus classification for the specimen; and
verify the predicted classification by a verification network trained solely on lipemia in response to the first hypothesizing network predicting a lipemia classification for the specimen.
20 . The non-transient computer-readable data storage medium of claim 16 , wherein the plurality of verification networks comprises two or more convolutional neural networks.
21 . A specimen testing apparatus, comprising:
a track;
a carrier moveable on the track and configured to contain a specimen container containing a serum or plasma portion of a specimen therein;
a plurality of image capture devices arranged around the track and operative to capture one or more images from one or more viewpoints of the specimen container and the serum or plasma portion of the specimen; and
the non-transient computer-readable data storage medium of claim 16 .