Urinary cytology artificial intelligence urinary tract epithelium cancer identification system

Abstract

The invention relates to a urinary cytology artificial intelligence urinary tract epithelium cancer identification system. The system includes: a data classification module used for classifying urinary cytology data acquired in advance to obtain a urinary cytology positive data set and a urinary cytology negative data set, and matching the urinary cytology positive data set and the urinary cytology negative data set based on an operation pathology result to obtain a true positive data set, a false positive data set, a true negative data set and a false negative data set; a data grouping module which is used for grouping various data sets obtained by the data classification module to obtain a training-verification set and a test set; a model training module which is used for training a pre-constructed model to obtain a final model; a model testing and auditing module which is used for performing cell level testing and tissue level testing on the obtained final model; and an identification module which is used for identifying the to-be-identified urinary cytology data to obtain an identification result. The system can be widely applied to the field of cytology pathology recognition.

Description

technical field [0001] The invention belongs to the field of medical technology, and relates to a urine cytology recognition system, in particular to a urine cytology artificial intelligence urothelial cancer recognition system based on deep learning. Background technique [0002] Urothelial carcinoma is the fourth most common malignancy worldwide. Since the renal pelvis, ureter, and bladder are all covered by the urothelium, urothelial carcinoma is often not limited to a single point but involves multiple places at the same time, and some urothelial carcinomas are prone to recurrence and progression after treatment. Therefore, at present, a thorough examination before treatment is advocated, and frequent review is also required after treatment. Surgical pathology is the gold standard for the diagnosis of urothelial carcinoma. However, due to the trauma and resource consumption caused by the operation, sufficient examination must be carried out before the operation, and it ...

AI Technical Summary

Problems solved by technology

Algorithmically, the image recognition algorithms of these systems need to be segmented and then classified, that is, researchers need to pre-define image features (such as cell nucleus edge, nucleus-to-cytoplasm ratio, cell volume, etc.), and then the algorithm first segments the original image and then captures the features. Therefore, the steps of the algorithm are increased and the cumulative error is increased, so this type of algorithm requires a large number of training samples (10^5-10^6 number of labeled cells)

On the other hand, due to the use of artificially defined image features, such algorithms are often fragile in practice and difficult to generalize, which does not conform to the current computer vision concept of relying on the algorithm itself to find features

Technically, the preparation of cytology in these studies relies on liquid-based thin-layer cytology (Thinprep cytologic test) and whole-field digital slide (Whole Slide Image) technology, and the Paris scoring system is used for diagnosis. However, the above-mentioned technology and scoring system Not yet widely available

Images