| CPC G16H 50/20 (2018.01) [G06T 7/0012 (2013.01); G06T 7/136 (2017.01); G16H 15/00 (2018.01); G16H 30/40 (2018.01); G06T 2207/10061 (2013.01); G06T 2207/20081 (2013.01); G06T 2207/30024 (2013.01)] | 6 Claims |
|
1. A method for an artificial intelligence cloud diagnosis platform, which comprising the following steps:
S1, numbering subject samples to determine sample numbers in a cloud system;
S2, registering so as to enter subject information into the system and enter the sample numbers;
scanning so as to digitalize the samples;
S3, uploading so as to upload the digitalized samples to the cloud system;
S4, stitching classification so as to process the digitalized samples on cloud AI;
S5, connecting so as to associate registration information with information of the digitalized sample in the system;
S6, diagnosing so as to diagnose and review the samples, and submit a diagnosis opinion operation by a doctor; and
S7, report rendering so as to poll the completely diagnosed data in the system by using a rendering program and rendering the data into PDF, JPG, WORD format files according to corresponding report templates thereof; wherein
auxiliary diagnosis on the cloud system is realized through above steps;
in step S4, a plurality of images of a single sample are stitched, wherein an image stitching process comprises: visual field sub-block matching, visual field position fitting and block extraction;
a process of the visual field sub-block matching is as follows:
Sa01, inputting and initiating a result set M;
Sa02, setting the current visual field i as a first visual field;
Sa03, solving a set J of all adjacent visual fields of the current visual field i;
Sa04, setting the current adjacent visual field j as a first visual field in J;
Sa05, solving possible overlapping regions Ri and Rj of the visual field i and the visual field j;
Sa06, rasterizing a template region Ri into template sub-block sets Pi;
Sa07, sorting the template sub-block sets Pi in a descending order according to a dynamic range of the sub-blocks;
Sa08, setting the current template sub-block P as the first one in the template sub-block sets Pi;
Sa09, solving a possible overlapping region s of the template sub-block P in the visual field J;
Sa10, performing a template matching search by taking the template sub-block P as a template and s as a search region;
Sa11, adding a best match m to the result set M;
Sa12, finding all matching visual field sets N that are in consistent with m from the result set M;
Sa13, judging whether or not a weight in N is greater than a threshold v upon comparison;
if not, setting the current template sub-block P as the next one in the template sub-block sets Pi and returning to Sa09;
if yes, proceeding to next step;
Sa14, judging whether or not the visual field j is the last visual field in the visual field set J upon comparison;
if not, setting the visual field j as the next visual field in the visual field set J and returning to Sa05;
if yes, proceeding to next step;
Sa15, judging whether or not the visual field i is the last visual field upon comparison;
if not, setting i as the next visual field and returning to Sa03;
if yes, outputting a result;
after the image stitching is completed by above steps, the stitched image is extracted according to features of a cell nucleus to acquire the microscopic images of the single cell nucleus;
a process of acquiring the microscopic images of the single cell nucleus is as follows:
Sa100, detecting features points of the cell nucleus;
reducing the image to a plurality of different scales and extracting feature points respectively;
Sa101, performing preliminary screening so as to screen to remove feature points that are too close by using coordinates of the feature points, thereby reducing repeated extraction of cells;
Sa102, subdividing so as to segment according to a color difference threshold;
converting a picture to a LAB format; and after an inversion of a B channel as well as the weighting and Otsu thresholding of an A channel, segmenting to acquire a cell nucleus mask map, wherein
the weight is 0.7 for the B channel under the inversion and 0.3 for the A channel;
S103, performing image morphology operation:
one or a combination of more of corrosion operation and expansion operation; and
S104, performing fine screening according to a nuclear occupancy parameter to remove non-cells each having a nuclear occupancy ratio below 0.3 and a nucleus radius above 150 pixels and below 10 pixels, wherein the nuclear occupancy ratio is obtained by dividing a nuclear area finely segmented according to the color difference threshold by a radius circle area of the detected feature point; and
after the microscopic images of the single cell nucleus are acquired by the above steps, the microscopic images of the single cell nucleus are classified according to the labeled cells by means of an artificial intelligence program subjected to model training;
thereby obtaining target-based classified cell data.
|