Podcast Debiao Li, PhD, Stephen Pandol, MD @Li_Debiao @CedarsSinai @csmcbiri #AI #Biomarker AI And Machine Learning Could Improve Cancer Dx

Debiao Li, Ph.D., Director, Biomedical Imaging Research Institute, Professor, Biomedical SciencesProfessor, Imaging, Principal Investigator at Cedars-Sinai, and Stephen Pandol, MD, Director, Basic and Translational Pancreas Research, Program Director, Gastroenterology Fellowship Program at Cedars-Sinai. In this video, they discuss the article AI and machine learning could improve cancer diagnosis through biomarker discovery. 

 

Many industries and fields of study have been changed by artificial intelligence (AI), deep learning (DL), and machine learning (ML). These techniques are now being used to solve the issues of cancer biomarker development, where huge volumes of imaging and molecular data must be analyzed in ways that classic statistical studies and tools cannot. Researchers propose numerous ways and discuss some of the particular problems of combining AI, DL, and ML to improve the accuracy and predictive value of biomarkers for cancer and other diseases in a special issue of Cancer Biomarkers.

 

Among the promising applications of AI, DL, and ML presented in this issue are identifying early-stage cancers, inferring the location of specific cancers, assisting in the assignment of appropriate therapeutic options for each patient, characterizing the tumor microenvironment, and predicting immunotherapy response.

 

A thorough review of the literature on the application of AI techniques to develop biomarkers for ovarian and pancreatic cancer illustrates basic principles and examines the gaps and issues that the field faces as a whole. Ovarian and pancreatic cancers are uncommon but deadly due to a lack of early symptoms and identification. Juergen A. Klenk, Ph.D., Biomedical Data Science Lab, Deloitte Consulting LLP, Arlington, VA, USA, and colleagues report experiments employing AI and ML to evaluate images for early disease diagnosis, as well as models that can be created to predict expected patient outcomes. Some of the difficulties are mentioned, such as the difficulty in acquiring large enough datasets.

 

To have a true influence on biomarker discovery, the researchers believe that larger and more diversified imaging datasets for uncommon tumors across institutions, consistent reporting techniques, and easier-to-understand interfaces that boost user confidence are required.

 

Debiao Li, Ph.D., of the Biomedical Imaging Research Institute at Cedars-Sinai Medical Center in Los Angeles, CA, and colleagues created a model to identify people at risk for pancreatic ductal adenocarcinoma (PDAC). PDAC is accompanied by several preconditional abnormalities that can be seen on computed tomography (CT) scans but are difficult to understand visually. The researchers identified a set of CT features that were possibly predictive of PDAC using CT scans from patients with confirmed PDAC and CT scans from the same patients who received a CT scan six months to three years before diagnosis. Using the detected CT features, the model was 86 percent accurate in categorizing the patients and healthy controls.

 

Radiomics is a new field in which features from medical imaging are extracted using various techniques. Radiomic characteristics, which can assess tumor intensity, shape, and heterogeneity, have been used in oncology detection, diagnosis, therapy response, and prognosis. Shaoli Song, Ph.D., Shanghai Medical College and Fudan University, Shanghai, China, and colleagues integrated radiomic data from preoperative positron emission tomography (PET) and CT imaging in patients with early stage uterine cervical squamous cell carcinoma. They created a prognostic signature capable of predicting disease-free survival using algorithms.