Image analysis method, image analysis apparatus, program, learned deep layer learning algorithm manufacturing method and learned deep layer learning algorithm

1. An image analysis method for analyzing an image of a tissue or a cell using a deep learning algorithm of a neural network structure, the method comprising:

generating analysis data from an analysis target image that includes the tissue or cell to be analyzed;

inputting the analysis data to the deep learning algorithm;

generating data indicating a tumorigenic state of a tissue or a cell in the analysis target image by the deep learning algorithm,

wherein the data indicating the tumorigenic state include, for each pixel of the analysis target image, label information generated by the deep learning algorithm of the neural network structure and respectively indicative of whether the each pixel of the analysis target image corresponds to a nuclear region of a tumor cell or to other regions, wherein the label information for each pixel is generated by performing a determination process for each pixel.

2. The image analysis method according to claim 1 , wherein

the analysis target image is an image of a tissue diagnostic sample, and the analysis target image includes a hue comprised of one primary color or a hue obtained by combining two or more primary colors.

3. The image analysis method according to claim 2 , wherein

the sample is a stained sample, and the analysis target image is an image obtained by imaging the stained sample under a bright field microscope.

4. The image analysis method according to claim 1 , wherein

the analysis target image is an image of a cell diagnostic sample, and the analysis target image includes a hue comprised of one primary color or a hue obtained by combining two or more primary colors.

5. The image analysis method according to claim 1 , wherein

the data indicating the tumorigenic state are data indicating a boundary between the nuclear region of the tumor cell and another region.

6. The image analysis method according to claim 1 , wherein

the deep learning algorithm determines whether a nucleus is a nucleus of a tumor cell or a nucleus of a non-tumor cell.

7. The image analysis method according to claim 1 , wherein

a plurality of the analysis data are generated for each region having a predetermined number of pixels relative to one analysis target image.

8. The image analysis method according to claim 7 , wherein

analysis data are generated for each region of the predetermined number of pixels including peripheral pixels circumscribing a predetermined pixel;

the deep learning algorithm generates a label indicating a tumorigenic state relative to the predetermined pixel of the input analysis data.

9. The image analysis method according to claim 7 , wherein

a number of nodes of an input layer of the neural network structure corresponds to the product of the predetermined number of pixels of the analysis data and a number of combined primary colors.

10. The image analysis method according to claim 1 , wherein training data used for learning of the deep learning algorithm are generated based on:

a sample image, which is a bright field image captured under a bright field microscope of a stained image of a sample prepared by staining for bright field observation of a sample of a tissue specimen containing a tumor collected from an individual or a sample of material containing a tumor cell collected from an individual; and

a fluorescence image of a nucleus of a cell nucleus corresponding to the sample or a stained image of a sample prepared by applying fluorescent nuclear stain to the same sample under fluorescence observation of a fluorescence microscope, the fluorescence image being such that the position in the sample of the fluorescence image corresponds a position in the sample of the obtained bright field image.

11. The image analysis method according to claim 10 , wherein

the staining for bright-field observation uses hematoxylin for nuclear staining.

12. The image analysis method of claim 11 , wherein

the stain for bright field observation is hematological-eosin stain when the sample is a tissue sample; and

the stain for bright field observation is Papanicolaou stain when the sample is a material containing cells.

13. The image analysis method according to claim 10 , wherein

the training data are generated for each region of a predetermined number of pixels in the bright field image.

14. The image analysis method according to claim 1 , wherein

training data used for learning of the deep learning algorithm include a label value indicating a tumorigenic state of cells extracted from a bright field image and a fluorescence image.

15. The image analysis method according to claim 14 , wherein

the training data include a label value indicative of tumorigenic state for each pixel of the bright field image.

16. The image analysis method according to claim 1 , wherein

the deep learning algorithm classifies the analysis data into classes indicating a tumorigenic state of cells contained in the analysis target image.

17. The image analysis method according to claim 1 , wherein

an output layer of the neural network structure is a node having a soft max function as an activation function.

18. The image analysis method according to claim 1 , further comprising:

outputting the data indicating the tumorigenic state of the tissue or the cell in the analysis target image generated by the deep learning algorithm.

19. The image analysis method according to claim 1 , wherein each label information is respectively indicative of whether the each pixel of the analysis target image corresponds to a nuclear region of a tumor cell, to a nuclear region of a non-tumor cell, or to other regions.

20. An image analysis apparatus for analyzing an image of a tissue or a cell using a deep learning algorithm of a neural network structure, the apparatus comprising:

a processing unit for generating analysis data from an analysis target image including an analysis target tissue or cell, inputting the analysis data into the deep learning algorithm, and generating data indicating the tumorigenic state of a tissue or a cell in an analysis target image by the deep learning algorithm,

wherein the data indicating the tumorigenic state include, for each pixel of the analysis target image, label information generated by the deep learning algorithm of the neural network structure and respectively indicative of whether the each pixel of the analysis target image corresponds to a nuclear region of a tumor cell or to other regions, wherein the processing unit is configured to generate the label information for each pixel by performing a determination process for each pixel.

21. The image analysis apparatus according to claim 20 , wherein each label information is respectively indicative of whether the each pixel of the analysis target image corresponds to a nuclear region of a tumor cell, to a nuclear region of a non-tumor cell, or to other regions.