This article, titled “Convergence of evolving artificial intelligence and machine learning techniques in precision oncology,” published in partnership with Seoul National University Bundang Hospital, explores the rapid advancements brought by Artificial Intelligence (AI) and Machine Learning (ML) in the field of precision oncology.
Authored by Elena Fountzilas, Tillman Pearce, Mehmet A. Baysal, Abhijit Chakraborty, and Apostolia M. Tsimberidou, the paper highlights how the confluence of new technologies with AI and ML analytical techniques is improving diagnostic approaches and therapeutic strategies for cancer patients. These technologies achieve this by analyzing multi-dimensional, multiomic, spatial pathology, and radiomic data, leading to a deeper understanding of complex molecular pathways and optimizing treatment selection.
The article details the extensive applications of AI/ML in precision oncology, including:
- Digital Pathology: AI/ML is being explored for automating immunohistochemistry (IHC) scoring, inferring clinically relevant features from hematoxylin and eosin (H&E) images, and gaining insights from emerging tools for measuring multiplex, single-cell, and spatially resolved analytes from tumor tissue. This includes automated PD-L1 evaluation and HER2 IHC scoring, which can standardize assessments, increase accuracy, and reduce turnaround time. Deep learning models have also shown the ability to predict molecular characteristics like HER2, BRCA expression, MSI/dMMR, EGFR, KRAS, and STK11 mutations directly from H&E-stained whole-slide images (WSIs).
- Digital Radiology (Radiomics): Radiomics involves high-throughput mining of quantitative image features from standard medical imaging to improve diagnostic, prognostic, and predictive accuracy. AI/ML algorithms can extract clinically relevant features from these datasets, providing insights not identifiable by traditional methods. Examples include predicting tumor-infiltrating lymphocyte (TIL) density and immunotherapy outcomes from CT imaging.
- Molecular Medicine: AI/ML tools are used to analyze “omics” data, such as next-generation sequencing (NGS), to identify novel biomarkers and drug targets. Specifically, CNNs have significantly improved variant calling accuracy in genomic analysis, outperforming existing tools. They also aid in analyzing epigenomic and proteomic datasets to identify patterns associated with tumor types, discover new epigenetic drugs, and develop predictive models.
- Integrative (Multimodal) Analyses: The paper emphasizes the development of multimodal AI models that incorporate various data sources like clinical, pathological, radiomic, and genomic characteristics, showing improved prediction of treatment responses compared to single-modality approaches.
- Large Language Models (LLMs) and Generative AI: Recent advances in deep learning, such as LLMs (e.g., GPTs), enable human interaction with computers using natural language and can generate text, images, or other data based on vast amounts of pre-trained data. LLMs show potential in facilitating decision support for cancer patients and mining electronic health records (EHRs) for relevant data.
Despite these promising applications, the article also addresses significant operational and technical challenges:
- Data Issues: These include difficulties with data technology, engineering, storage, quality, quantity, sharing, and generalizability. The need for large, diverse, and well-annotated datasets is crucial. Federated learning is presented as a transformative solution to enable multi-institutional collaboration while preserving patient privacy.
- Algorithm Development: Challenges exist in algorithm development and structures, particularly with complex models that often lack interpretability.
- Clinical Integration: Incorporating these technologies into the current clinical workflow and reimbursement models poses significant hurdles. Resistance to change, the need for extensive physician training, and the high costs associated with digital pathology transformation are also highlighted.
- Ethical and Regulatory Aspects: Concerns include data bias, lack of model transparency (the “black box” nature), issues of accuracy and reliability, accountability, and patient data privacy. Explainable AI (XAI) methods are crucial for building trust.
The paper notes that as of December 20, 2024, the FDA has approved 1016 AI/ML-enabled medical devices, with specific examples demonstrating successful clinical outcomes in breast cancer detection, skin cancer classification, prostate cancer detection, and lung cancer diagnosis.
In conclusion, while AI/ML tools hold immense promise for transforming precision oncology by optimizing treatment strategies and accelerating drug development, their widespread adoption requires rigorous clinical validation, standardization, ethical guidelines, and seamless integration into clinical practice.
Reference: Fountzilas, E., Pearce, T., Baysal, M. A., Chakraborty, A., & Tsimberidou, A. M. (2025). Convergence of evolving artificial intelligence and machine learning techniques in precision oncology. NPJ Precision Oncology, 25, Article 01471. https://doi.org/10.1038/s41746-025-01471-y
