- 건국대학교

	KU Research Team Develops Artificial Intelligence Cancer Diagnosis Platform
	관리자		406		2021.01.05

A research team of Konkuk University’s Medical School (Professor Kyungsik Park from College of Surgeons and Professor Sungyoung Kim from College of Biochemistry) announced that they have succeeded in the development of ‘Artificial Intelligence Cancer Diagnosis Platform’ that increases the consistency and interpretation of models using their own machine learning-based meta-analysis algorithms. <img style="margin-left: auto; display: block; margin-right: auto;" alt="" src="ArticleFile.do?id=11810434" /> Recently, there has been increasing interest in the field of artificial intelligence in medicine, and multiomics data and machine learning are expected to be innovative future medical technologies for cancer diagnosis and treatment. However, as predictive factors and models vary by researchers, centers and analysis platforms, the demand for discovery and verification of more objective and reliable models is highly required.
Meta-analysis is a major statistical technique in evidence-driven medicine as it integrates statistics of data from independent yet similar studies to evaluate the consistency of results and increase statistical accuracy. The research team has established even stronger model by combining meta-analysis with machine-learning algorithm based on their own biological pathway. The research team has found out that the merging of multiple cohorts and the dimensionality reduction method using individual biological pathways significantly increased the generality and metastasis of the learning model. The model in deed classified several subtypes of thyroid cancer nearly perfectly in the female cohort. The research team also found out the modulators of the pathway through analysis of key thyroid cancer-related biological pathways and multiomics analysis.
The newly found core biological pathway related to thyroid cancer by artificial intelligence is expected to provide a fundamental clue to the development of new thyroid cancer drugs. The developed algorithm, unlike other machine learning algorithms that are easily applicable to other cancers and difficult to interpret results such as deep learning, is expected to offer clinically preferred artificial intelligence solutions that value the cause analysis and consistency of results with regression and path analysis-based algorithms.
The research result was published in the latest issue of BIB (Briefings in Bioinformatics, IF:8.99), an international journal of mathematics and computer biology. The study was jointly conducted with Memorial Sloan-Kettering Cancer Center, the famous artificial intelligence cancer center, and was supported by Korea Research Foundation’s basic research projects in the field of science and technology and biomedical technology development. <img style="margin-left: auto; display: block; margin-right: auto;" alt="" src="ArticleFile.do?id=11810433" width="632" height="389" />▲ (Fig. 1) A schematic diagram of the developed meta-analysis-based machine learning algorithm As a core process of meta-analysis-based machine learning algorithm, 1) standardized multiple cohorts are mapped to individual biological pathways using nonlinear principal component analysis, 2) penalty-based machine learning and parameter optimization algorithms are applied, and 3) core predictors are extracted.


	Prof. Man-Jong Lee Reported a...
	Prof. Man-Jong Lee Reported t...