Development and multi-institutional validation of an artificial intelligence-based diagnostic system for gastric...

Area of Science:

• Digital pathology
• Artificial intelligence in medicine
• Computational pathology

Background:

• Pathologist workload for gastric biopsy diagnosis is increasing.
• Need for efficient and accurate diagnostic tools in pathology.

Purpose of the Study:

• Develop an artificial intelligence-based system for pathological diagnosis of gastric biopsies (AI-G).
• Evaluate the performance of the novel multistage semantic segmentation for pathology (MSP) method in AI-G.
• Assess the accuracy and robustness of MSP AI-G across multiple institutions.

Main Methods:

• Developed a novel multistage semantic segmentation for pathology (MSP) method for whole-slide image (WSI) analysis.
• Trained the AI-G system on 4511 gastric biopsy WSIs from 984 patients.
• Validated MSP AI-G performance on 3450 samples from 1772 patients across 10 different institutes.

Main Results:

• MSP AI-G achieved higher tissue-level accuracy (91.0%) compared to conventional patch-based AI-G (PB AI-G) (89.8%).
• MSP AI-G demonstrated superior accuracy (0.946 ± 0.023) versus PB AI-G (0.861 ± 0.078) in multi-institutional analysis.
• The system showed high diagnostic accuracy and robustness across diverse clinical settings.

Conclusions:

• The MSP AI-G system offers high diagnostic accuracy and robustness for gastric biopsies in multi-institutional settings.
• This AI-G system can potentially reduce the need for secondary review processes by assisting pathologists.
• The developed AI-G system shows promise for integration into daily clinical practice.