AI: A Useful Tool for Oncologists to Analyze Tumor Microenvironments?
By Michelle Xu
It is no secret that the usage of artificial intelligence has been dramatically popularized in recent years. From Chat-GPT to self-driving cars, the advancement of deep learning algorithms is rapidly growing. At the same time, cancer has become an increasingly prevalent issue for society to solve. Due to the availability of statistics regarding cancer, researchers have been utilizing artificial intelligence to help with cancer diagnosis, prognosis prediction, and treatment assignment.
One type of artificial intelligence that oncologists use to aid cancer diagnosis, prognosis prediction, or treatment decisions is deep learning models. Deep learning is a powerful tool for oncologists to extract features within the tumor microenvironment because of its ability to recognize patterns within large amounts of data, identify characteristics, and understand the data’s relationships. These skills are what allow artificial intelligence to be able to analyze images of tumor tissues and provide critical information about them.
Examples of features deep learning can extract within cancer tumors include the number of T cells in the tumor, the number of cancer cells, the composition of cancer cells and T cells, and more. This information is crucial when finding the appropriate treatment for a cancer patient. For instance, patients with high numbers of T cells in their tumor may be able to use immunotherapy and activate their T cells to fight the cancerous cells within the tumor. However, a patient with low numbers of T cells in their tumor may need other forms of treatment, such as chemotherapy or radiation therapy. Thus, the information deep learning provides regarding tumor microenvironments is essential for oncologists to decide the next steps in a patient’s treatment process.
Despite its advantages, the usage of deep learning to study tumor microenvironments comes with several limitations. For deep learning models to be the most effective, large sets of data are needed to train the deep learning models and maximize their performance. Unfortunately, high-quality tumor datasets are difficult and expensive to obtain. Because of the heterogeneity of cancer as a disease, large datasets are crucial for the accuracy of deep learning models. Additionally, uncertainty plays a fundamental role in deep learning models’ ability to make predictions and analyze tumor microenvironments. If certainty levels are too low, artificial intelligence can provide inaccurate predictions and information. In most severe cases, this could be harmful and possibly fatal for a patient battling cancer.
Despite the challenges that deep learning poses, artificial intelligence has proved to be a successful and accurate tool in the analysis of tumor microenvironments. By allocating more resources to the development and training of deep learning models within the study of tumor microenvironments, oncologists may be able to effectively solve those challenges and use artificial intelligence to save a patient’s life.
Works Cited
Kumar Y, Gupta S, Singla R, Hu YC. A Systematic Review of Artificial Intelligence Techniques in Cancer Prediction and Diagnosis. Arch Comput Methods Eng. 2022;29(4):2043-2070. doi: 10.1007/s11831-021-09648-w. Epub 2021 Sep 27. PMID: 34602811; PMCID: PMC8475374.
Tran, K.A., Kondrashova, O., Bradley, A., et al. Deep learning in cancer diagnosis, prognosis, and treatment selection. Genome Med 13, 152 (2021). https://doi.org/10.1186/s13073-021-00968-x
