AI and machine learning could improve cancer diagnosis through biomarker discovery

Reviewed by Emily Henderson, B.Sc.Mar 1 2022

Artificial intelligence (AI), deep learning (DL), and machine learning (ML) have transformed many industries and areas of science. Now, these tools are being applied to address the challenges of cancer biomarker discovery, where the analysis of vast amounts of imaging and molecular data is beyond the ability of traditional statistical analyses and tools. In a special issue of Cancer Biomarkers, researchers propose various approaches and explore some of the unique challenges of using AI, DL, and ML to improve the accuracy and predictive power of biomarkers for cancer and other diseases.

"The biomarker field is blessed with a plethora of imaging and molecular-based data, and at the same time, plagued with so much data that no one individual can comprehend it all," explained Guest Editor Karin Rodland, PhD, Pacific Northwest National Laboratory, Richland; and Oregon Health and Science University, Portland, OR, USA. "AI offers a solution to that problem, and it has the potential to uncover novel interactions that more accurately reflect the biology of cancer and other diseases."

Promising applications of AI, DL, and ML presented in this issue include identifying early-stage cancers, inferring the site of the specific cancer, aiding in the assignment of appropriate therapeutic options for each patient, characterizing the tumor microenvironment, and predicting the response to immunotherapy.

A comprehensive overview of the literature regarding the use of AI approaches to identify biomarkers for ovarian and pancreatic cancer illustrates underlying principles and looks at the gaps and challenges that face the field as a whole. Ovarian and pancreatic cancers are rare, but lethal because they lack early symptoms and detection. Lead investigator Juergen A. Klenk, PhD, Biomedical Data Science Lab, Deloitte Consulting LLP, Arlington, VA, USA, and colleagues describe studies using AI and ML to analyze images for the early detection of disease, and models that can be built to predict likely outcomes for the patient. Some of the challenges, such as the difficulty of gathering large enough datasets, are discussed.

Algorithms develop biases and produce prejudiced responses when the data they are trained on are non-representative or incomplete."

Dr. Juergen A. Klenk, PhD, Lead Investigator, Biomedical Data Science Lab, Deloitte Consulting LLP, Arlington, VA, USA

The investigators suggest that the development of larger and more diverse image databases for rare cancers across institutions, standardized reporting methods, and easier-to-understand interfaces that increase user trust are needed to make a true impact on biomarker discovery.

Lead investigator Debiao Li, PhD, Biomedical Imaging Research Institute, Cedars-Sinai Medical Center, Los Angeles, CA, USA, and colleagues developed a model to identify individuals at risk for pancreatic ductal adenocarcinoma (PDAC). PDAC is associated with many preconditional abnormalities that can be visible on a computerized tomography (CT) scan, but these are difficult to comprehend by visual assessment. In their study, the investigators used CT scans from patients with confirmed PDAC and CT scans from the same patients who had had a CT scan six months to three years before diagnosis to identify a set of CT features that were potentially predictive of PDAC. The model was 86% accurate in classifying the patients and the healthy controls, using the identified CT features.

"The challenge of AI for the advancement of pancreatic cancer research is the scarcity of data due to low prevalence. The purpose of this proof-of concept model Is to encourage researchers to establish a larger dataset for extensive training and validation of the model," said Dr. Li.

Radiomics is an emerging field where features are extracted from medical imaging using various techniques. Radiomic features can quantify tumor intensity, shape, and heterogeneity and have been applied to oncologic detection, diagnosis, therapeutic response, and prognosis. Lead investigators Shaoli Song, PhD, Shanghai Medical College and Fudan University, Shanghai, China, and Lisheng Wang, PhD, Shanghai Jiao Tong University, Shanghai, China, and colleagues combined radiomic data from preoperative positron emission tomography (PET) and CT images in patients with early stage uterine cervical squamous cell carcinoma. They used algorithms to develop a prognostic signature capable of predicting disease-free survival.

"This model could provide more accurate information about potential relapse and metastasis, and could be helpful in decision-making," they observed.

Other papers in the special issue focus on the development of new computational tools to facilitate the application of AI to biomarker identification; the use of whole cell imaging and immunofluorescence to identify immune features in pancreatic tumors to provide prognostic information; the use of microRNAs and applied machine learning to identify a miRNA profile associated with gastrointestinal stromal tumors; and the use of hierarchical clustering of combined multi-omic datasets to identify an antitumor immune signature in patients with colon cancer.

Dr. Rodland added that the articles in this special issue are only a small sampling of the various approaches to using AI, DL, and ML in biomarker research. "There is a continuing urgent need for more effective strategies for improving the early detection of cancers. Cutting-edge AI systems have been shown to improve sensitivity and specificity in the interpretation of both imaging and non-imaging data for breast, lung, prostate, and cervical cancers," she stated.

Noted experts comment on the special issue

Anirban Maitra, MBBS, MD Anderson Cancer Center
As the universe of cancer research and clinical care expands with the generation of ever larger datasets and integration of data across diverse platforms, it comes as no surprise that AI and ML are increasingly being adopted into oncology. For those of us familiar with the unfortunate phenomenon of "missed cancers" on serial imaging scans or biomarker assays, especially in high-risk individuals, AI/ML-based tools can be pivotal. This issue is highly timely and provides a sampling of the excitement permeating the field.

Kenneth W. Kinzler, PhD, Johns Hopkins Kimmel Cancer Center
Advances in machine learning are impinging on our daily lives in an ever-increasing manner. The same is true in biomedical research, especially in the area of cancer research where ML approaches are promising to improve our ability to detect cancer early and enhance patient management. This special issue demonstrates the ability of ML to improve cancer research in areas as diverse as early detection and electronic medical records.

Chris Amos, PhD, Baylor College of Medicine
This special issue brings together a wealth of new approaches for applying new technologies in machine learning and artificial intelligence with advances in high-throughput biomarker analysis to characterize patterns that identify individuals at high risk for developing cancer. It provides a great resource for computational scientists, researchers, and clinicians to understand these state-of-the-art developments.

Samir M. Hanash, MD, PhD, MD Anderson Cancer Center
Current interest in biomarkers spans the need for personalized cancer therapy and monitoring for disease progression and recurrence to cancer risk assessment and early detection. There is a wide world of platforms for biomarker discovery from genomics to proteomics and metabolomics, among others, that yield vast amounts of data that benefit from AI approaches to data analysis. This special issue is timely as it addresses the application of AI to cancer research and the contribution of AI for improving cancer detection and diagnosis through biomarker discovery.