Related Experiment Video
Updated: Jun 16, 2025
A Rapid Method for Multispectral Fluorescence Imaging of Frozen Tissue Sections
Published on: March 30, 2020
Perovskite Probe-Based Machine Learning Imaging Model for Rapid Pathologic Diagnosis of Cancers
Jimei Chi1,2, Yonggan Xue3, Yinying Zhou4
1Key Laboratory of Green Printing, CAS Research/Education Center for Excellence in Molecular Sciences, Institute of Chemistry, Chinese Academy of Sciences (ICCAS), Beijing Engineering Research Center of Nanomaterials for Green Printing Technology, Beijing National Laboratory for Molecular Sciences (BNLMS), Beijing 100190, P. R. China.
This study introduces a rapid, machine learning-powered imaging method using perovskite nanocrystal probes for accurate cancer diagnosis. The advanced probe significantly improves tumor detection, aiding in better surgical outcomes and reduced mortality.
Area of Science:
- Biomedical Imaging
- Nanotechnology
- Machine Learning in Pathology
Background:
- Accurate tumor cell identification is crucial for cancer diagnosis, staging, and treatment.
- Conventional fluorescence immunohistochemistry struggles with tumor cell heterogeneity and lacks big data analysis capabilities.
- Existing methods face challenges in rapid, precise differentiation of cancerous from normal tissues.
Purpose of the Study:
- To develop a machine learning-driven imaging method for rapid pathological diagnosis of five major cancer types.
- To utilize high-efficiency perovskite nanocrystal probes for enhanced cancer detection at the single-cell level.
- To improve the accuracy of surgical resection and reduce cancer mortality through precise diagnosis.
Main Methods:
- Developed perovskite nanocrystal probes modified with survivin antibodies for cancer bioanalysis.
- Employed machine learning to analyze fluorescence intensity and pathological texture from 1000 images.
- Validated the method for rapid differentiation of tumor from normal tissues within 10 minutes.
Main Results:
- Achieved a 10.3-fold higher tumor-to-normal (T/N) ratio compared to conventional probes.
- Machine learning classification achieved an area under the curve >90% for five cancer types (breast, colon, liver, lung, stomach).
- Successfully predicted the tumor organ in 92% of positive patients.
Conclusions:
- The developed method offers a high T/N ratio probe for precise, multi-cancer diagnosis.
- Machine learning integration enhances diagnostic accuracy and speed.
- This approach holds significant potential for improving surgical precision and lowering cancer mortality rates.
