Page 115 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

(continued)

Item Details
Data Availabil- Private
ity 4.1-Healthcare
Metadata (Type Image, Electronic Medical Record (EMR) Report, Spatiotemporal Omics
of Data)

Model Training • Mixture of Vision Encoder with Self-Supervised Learning (SSL)
and Fine-Tun- • Vision-Language Alignment and Vision-Language Model Supervised
ing Fine-Tuning (VLM SFT)

Testbeds or Deployed internally at Ruijin Hospital [2], [3].
Pilot Deploy-
ments

Code reposito- Not yet available. A GitHub link will be added upon release.
ries

2 Use Case Description

2�1 Description

According to the latest report released by the World Health Organization's International
Agency for Research on Cancer (IARC), the global cancer burden is increasing, with 20 million
new cancer cases and 9.7 million deaths worldwide by 2022. Early detection, early diagnosis
and early treatment are the key. It is urgent to expand the accessibility of pathology diagnosis,
improve the accuracy of pathology diagnosis and improve the level of pathology diagnosis
at the grass-roots level. For example, in China, the following problems exist in the pathology
industry in the medical field: (1) There is a large gap in the number of doctors: no more
than 20,000 registered pathologists, with a gap of 70,000-140,000. (2) Uneven distribution of
pathologists: 70% are concentrated in third-class hospitals. (3) Low compliance rate of initial
diagnosis: The complete compliance rate of initial diagnosis opinions in primary hospitals is
only 13%, and the general compliance rate of initial diagnosis opinions is about 30%.

The DCS AI full-stack data storage solution has built the first multi-modality pathology model in
China for automatic generation of clinical pathology reports. Pixel-level comparison, full slice
traversal, interactive AI-assisted reading, and pathology report generation time were shortened
from 40 minutes to 15 minutes, improving efficiency by 75%. From traditional microscopes
to digitalization to intelligence, AI technology is used to achieve intelligent consultation and
precise screening, improve the early diagnosis rate and treatment efficiency of common
diseases, reduce misdiagnosis rates, help make up for the shortage of medical resources,
align with international standards, and improve global public health standards.

In DCS AI full-stack data storage solution, the AI-Enabled Pathology Model named RuiPath model
represents an innovative multimodal AI system designed for pathology analysis, integrating
advanced machine learning techniques with large-scale medical data. Its technological
approach was characterized by the following aspects:

79