Pattern recognition method of autoantibody immunofluorescence image
A pattern recognition method of the immunofluorescence images of autoantibody identification is disclosed. The method includes the following steps: inputting a plurality of original cell immunofluorescence images; conducting an operation of a plurality of convolutional neural networks by a processor, the plurality of convolutional neural networks include a convolution layer, a pooling layer and an inception layer for capturing the plurality of convolution features; conducting a judgment process to obtain the proportions of the antinuclear antibodies morphological patterns; and outputting the recognition results.
1. A pattern recognition method of immunofluorescence images of autoantibody identification, comprising the following steps:
inputting a plurality of original cell immunofluorescence images through an input device, converting the plurality of original cell immunofluorescence images into a plurality of three primary color layers, respectively;
conducting an operation of a plurality of convolutional neural networks on the plurality of three primary color layers via a processor, wherein each of the plurality of convolutional neural networks comprises a convolution layer, a pooling layer and an inception layer for capturing a plurality of convolution features after the operation, followed by using the plurality of convolution features as input of next order of the convolutional neural networks;
conducting a recognition process via the processor, fully connecting the plurality of convolution features with a plurality of neurons, recognizing proportions of the plurality of original cell immunofluorescence images and comparing the proportions of the plurality of original cell immunofluorescence images to a plurality of antinuclear antibody patterns; and
outputting recognition results of the plurality of original cell immunofluorescence images through an output device.
2. The pattern recognition method according to claim 1 , wherein the convolution layer comprises a trigger function, and the trigger function comprises a Sigmoid function, a Tan h function, a ReLU function, a PReLU function, an ArcTan function, an ELU function, a SoftPlus function, a Sinusoid function, a Sinc function, a Bent identity function, or a Gaussian function.
3. The pattern recognition method according to claim 1 , wherein the pooling layer comprises a max-pooling operation or a mean pool operation.
4. The pattern recognition method according to claim 1 , wherein the inception layer comprises a convolution operation and a max-pooling operation for different sizes of templates in a same layer of operation, and then data are cascaded.
5. The pattern recognition method according to claim 1 , wherein each of the plurality of antinuclear antibody patterns comprises a phenotype selected from the group consisting of: a negative phenotype, a nuclear phenotype, a cytoplasmic phenotype, a mitotic phenotype, and a combination thereof.
6. The pattern recognition method according to claim 5 , wherein the nuclear phenotype comprises a homogeneous pattern, a speckled pattern, a centromere pattern, a discrete nuclear dots pattern, a nucleolar pattern, a nuclear envelope pattern, and a pleomorphic pattern.
7. The pattern recognition method according to claim 6 , wherein the speckled pattern includes a dense fine speckled pattern, a fine speckled pattern, a large coarse speckled pattern, and a Topo I-like pattern.
8. The pattern recognition method according to claim 6 , wherein the discrete nuclear dots pattern includes a multiple nuclear dots pattern and a few nuclear dots pattern.
9. The pattern recognition method according to claim 6 , wherein the nucleolar pattern includes a homogeneous pattern, a clumpy pattern, and a punctate pattern.
10. The pattern recognition method according to claim 6 , wherein the nuclear envelope pattern includes a smooth pattern and a punctate pattern.
11. The pattern recognition method according to claim 6 , wherein the pleomorphic pattern includes a PCNA pattern and a CENP-F pattern.
12. The pattern recognition method according to claim 5 , wherein the cytoplasmic phenotype comprises a fibrillar pattern, a speckled pattern, an anti-mitochondrial antibody (AMA) pattern, a Golgi pattern, and a rods and rings pattern.
13. The pattern recognition method according to claim 12 , wherein the fibrillar pattern includes a linear pattern, a filamentous pattern, and a segmental pattern.
14. The pattern recognition method according to claim 12 , wherein the speckled pattern includes a discrete dots pattern, a dense fine speckled pattern, and a fine speckled pattern.
15. The pattern recognition method according to claim 5 , wherein the mitotic phenotype comprises a centrosome pattern, a spindle fibers pattern, an intercellular bridge pattern, and a mitotic chromosomal envelope pattern.
16. The pattern recognition method according to claim 15 , wherein the spindle fibers pattern includes a nuclear mitotic apparatus (NuMA) pattern.
17. The pattern recognition method according to claim 1 , wherein the plurality of original cell immunofluorescence images inputted are subjected to an operation of 10-layer convolutional neural networks to capture the plurality of convolution features.
18. The pattern recognition method according to claim 1 , wherein the convolution layer in each of the plurality of convolutional neural networks has a convolution kernel of a predetermined pixel size.