Methods and apparatus for fine-grained HIL index determination with advanced semantic segmentation and adversarial training
A method of characterizing a serum or plasma portion of a specimen in a specimen container provides a fine-grained HILN index (hemolysis, icterus, lipemia, normal) of the serum or plasma portion of the specimen, wherein the H, I, and L classes may each have five to seven sub-classes. The HILN index may also have one un-centrifuged class. Pixel data of an input image of the specimen container may be processed by a deep semantic segmentation network having, in some embodiments, more than 100 layers. A small front-end container segmentation network may be used to determine a container type and boundary, which may additionally be input to the deep semantic segmentation network. A discriminative network may be used to train the deep semantic segmentation network to generate a homogeneously structured output. Quality check modules and testing apparatus configured to carry out the method are also described, as are other aspects.
1. A method of characterizing a specimen container, comprising:
capturing an image of the specimen container with an image capture device, the specimen container including a serum or plasma portion of a specimen therein;
training a deep semantic segmentation network with a discriminative network such that the deep semantic segmentation network generates a homogeneously structured output;
processing pixel data of the image of the specimen container using the deep semantic segmentation network executing on a computer to determine a classification index of the serum or plasma portion; and
outputting from the deep semantic segmentation network the classification index of the serum or plasma portion, wherein the classification index comprises hemolytic, icteric, lipemic, and normal classes, and each of the hemolytic, icteric, and lipemic classes comprises five to seven sub-classes.
2. The method of claim 1 , wherein the classification index further comprises an un-centrifuged class.
3. The method of claim 1 , wherein the processing pixel data comprises determining the classification index of the serum or plasma portion based on pixel color data from the image of the serum or plasma portion identified in the specimen container using the deep semantic segmentation network executing on the computer.
4. The method of claim 1 , further comprising processing the pixel data of the image of the specimen container using a front-end container segmentation network executing on the computer to determine a container type and a container boundary.
5. The method of claim 1 , wherein the deep semantic segmentation network comprises an architecture including at least ten dense block layers.
6. A method of characterizing a specimen container, comprising:
capturing an image of the specimen container with an image capture device, the specimen container including a serum or plasma portion of a specimen therein;
processing pixel data of the image of the specimen container using a front-end container segmentation network executing on a computer to determine a container type and a container boundary;
processing the pixel data of the image of the specimen container using a deep semantic segmentation network executing on the computer to determine a classification index of the serum or plasma portion;
inputting the container type and container boundary from the front-end container segmentation network via an additional input channel to the deep semantic segmentation network to determine the classification index of the serum or plasma portion; and
outputting from the deep semantic segmentation network the classification index of the serum or plasma portion, wherein the classification index comprises hemolytic, icteric, lipemic, and normal classes, and each of the hemolytic, icteric, and lipemic classes comprises five to seven sub-classes.
7. The method of claim 6 , further comprising training the deep semantic segmentation network with a discriminative network such that the deep semantic segmentation network generates a homogeneously structured output.
8. A method of characterizing a specimen container, comprising:
capturing an image of the specimen container with an image capture device, the specimen container including a serum or plasma portion of a specimen therein;
processing pixel data of the image of the specimen container using a deep semantic segmentation network executing on a computer to determine a classification index of the serum or plasma portion; and
outputting from the deep semantic segmentation network the classification index of the serum or plasma portion, wherein the classification index comprises hemolytic, icteric, lipemic, and normal classes, and each of the hemolytic, icteric, and lipemic classes comprises five to seven sub-classes; wherein:
the deep semantic segmentation network comprises an architecture including at least ten dense block layers; and
the at least ten dense block layers includes at least three dense block layers each having a maximum of four dense layers and at least two dense block layers each having a maximum of five dense layers.
9. A method of characterizing a specimen container, comprising:
capturing an image of the specimen container with an image capture device, the specimen container including a serum or plasma portion of a specimen therein;
processing pixel data of the image of the specimen container using a deep semantic segmentation network executing on a computer to determine a classification index of the serum or plasma portion; and
outputting from the deep semantic segmentation network the classification index of the serum or plasma portion, wherein the classification index comprises hemolytic, icteric, lipemic, and normal classes, and each of the hemolytic, icteric, and lipemic classes comprises five to seven sub-classes; wherein:
the deep semantic segmentation network comprises an architecture including at least ten dense block layers; and
the deep semantic segmentation network comprises an architecture including at least one dense block layer having 15 dense layers.
10. A method of characterizing a specimen container, comprising:
capturing an image of the specimen container with an image capture device, the specimen container including a serum or plasma portion of a specimen therein;
processing pixel data of the image of the specimen container using a deep semantic segmentation network executing on a computer to determine a classification index of the serum or plasma portion; and
outputting from the deep semantic segmentation network the classification index of the serum or plasma portion, wherein the classification index comprises hemolytic, icteric, lipemic, and normal classes, and each of the hemolytic, icteric, and lipemic classes comprises five to seven sub-classes; wherein:
the deep semantic segmentation network comprises an architecture including at least ten dense block layers; and
the deep semantic segmentation network comprises an architecture including five transition down layers followed by five transition up layers.
11. A quality check module, comprising:
a plurality of image capture devices operative to capture one or more images from one or more viewpoints of a specimen container containing a serum or plasma portion of a specimen therein; and
a computer coupled to the plurality of image capture devices, the computer configured and operative to:
train a deep semantic segmentation network with a discriminative network such that the deep semantic segmentation network generates a homogeneously structured output;
process pixel data of the one or more images of the specimen container using the deep semantic segmentation network executing on the computer to determine a classification index of the serum or plasma portion; and
output from the deep semantic segmentation network the classification index of the serum or plasma portion, wherein the classification index comprises hemolytic, icteric, lipemic, and normal classes, and each of the hemolytic, icteric, and lipemic classes comprises five to seven sub-classes.
12. The quality check module of claim 11 , wherein the computer is further configured and operative to output the classification index further comprising an un-centrifuged class.
13. The quality check module of claim 11 , wherein the computer is further configured and operative to process the pixel data of the one or more images using a front-end container segmentation network executing on the computer to determine a container type and a container boundary.
14. 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; and
the quality check module of claim 11 coupled to the track and operative to receive the specimen container via the track.
15. A quality check module comprising:
a plurality of image capture devices operative to capture one or more images from one or more viewpoints of a specimen container containing a serum or plasma portion of a specimen therein; and
a computer coupled to the plurality of image capture devices, the computer configured and operative to:
process pixel data of the one or more images using a front-end container segmentation network executing on the computer to determine a container type and a container boundary; process the pixel data of the one or more images of the specimen container using a deep semantic segmentation network executing on the computer to determine a classification index of the serum or plasma portion;
input the container type and the container boundary from the front-end container segmentation network via an additional input channel to the deep semantic segmentation network to determine the classification index of the serum or plasma portion; and
output from the deep semantic segmentation network the classification index of the serum or plasma portion, wherein the classification index comprises hemolytic, icteric, lipemic, and normal classes, and each of the hemolytic, icteric, and lipemic classes comprises five to seven sub-classes.
16. The quality check module of claim 15 , wherein the computer is further configured and operative to train the deep semantic segmentation network with a discriminative network such that the deep semantic segmentation network generates a homogeneously structured output.
17. 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; and
the quality check module of claim 14 coupled to the track and operative to receive the specimen container via the track.
18. The quality check module of claim 15 , wherein the computer is further configured and operative to output the classification index further comprising an un-centrifuged class.