Shi Huaiyin: Application of AI in Pathological Diagnosis of Digestive Tract

Application of AI in Pathology

1) Accurate acquisition of diseased tissue;

2) Pathological diagnosis;

3) Histological grading and quantitative scoring;

4) Evaluation of tumor biomarkers;

5) Predict the molecular features of tumors based on HE images;

6) Predict the prognosis of patients based on HE images;

7) Information integration enables deep and accurate diagnosis
.

Necessity for the development of artificial intelligence auxiliary systems for pathological diagnosis: the shortage of pathological diagnosis talents and the advantages of AI in some fields of pathological diagnosis
.

The development of digital pathology has laid the foundation for the research and development of AI-assisted pathological diagnosis systems
.

The AI-assisted pathological diagnosis platform needs to meet the following conditions

1) high accuracy;

2) Compared with the traditional section microscope diagnosis, the diagnosis time is significantly shortened or at least not prolonged;

3) The application scenario is simple and convenient, and no additional operation steps are required;

4) The stability of the system when running in different hospital pathology departments

Current research and development and application status of AI in pathological diagnosis

1) Cervical cytology AI-assisted pathological diagnosis;

2) AI-assisted pathological diagnosis of gastrointestinal endoscopy biopsy specimens;

3) AI-assisted gross and microscopic image restoration of ESD specimens;

4) AI-assisted pathological diagnosis of lymph node metastases;

5) AI-assisted pathological diagnosis of breast cancer;

6) AI-assisted pathological diagnosis of lung cancer;

7) AI-assisted diagnosis of the positive degree of biomarkers such as KI-67, HER-2, PD-L1;

8) Predicting tumor molecular biological characteristics such as EGFR mutation of lung cancer, AI-assisted pathological diagnosis,
etc.

The first AI diagnosis model for gastrointestinal pathology that has entered practical clinical application and its advantages: greatly shorten the time compared with the traditional gastrointestinal endoscopy biopsy specimen diagnosis process

Diagnostic example:

Problems existing in AI-assisted pathological diagnosis system for gastrointestinal mucosal biopsy specimens

1) It is too sensitive and the false positive rate is too high;

2) There is no clear standard for the judgment of atrophic gastritis and its degree;

3) There is no clear standard for judging the degree of glandular hyperplasia;

4) HP identification is difficult
.

The product operation process of the ESD specimen visualization diagnostic system:

Challenges faced by pathological diagnosis AI

Challenge 1 - The volume of pathological images is huge, and storage is a big problem;

Challenge 2 - Diversification of specimen types and staining types (histological specimens, general HE staining, special staining, immunohistochemical staining, FISH fluorescence textual hybridization, etc.
)

Challenge 3 - Pathomorphological Diversity

Challenge 4 - There is a lot of content to learn (take a large specimen as an example: including histological type, histological grade, identification of necrosis, identification of vascular tumor thrombus, neural invasion, depth of invasion, presence or absence of lymph node metastasis, and related immune groups) Interpretation of chemical staining results, comprehensive analysis and judgment of immunohistochemical staining results and histological morphology, and comprehensive analysis of clinical information)
.

Challenge 5 - Complexity in quality control, equipment, technology (quality control of HE slides and immunohistochemically stained slides, popularity of slide scanners, increased speed of slide scanning, large storage capacity, digital slides and PACS systems integration of AI-assisted diagnosis system with HIS and PACS systems, automatic classification of different types of digital slices)
.

Thinking and Countermeasures

1) Objectively understand the application of AI in pathological diagnosis: to assist in pathological diagnosis, the expectations and requirements cannot be too high;

2) The correct leadership of the pathology industry association: priority field selection, learning content design, expert team demonstration, AI product development, and AI product multi-center verification
.

The future of smart pathology

In short, the future of intelligent pathology is still worth looking forward to
.
Highly intelligent and cloud-based is the goal of the future
.