Significant advancements in understanding the molecular and cellular mechanisms of tumor progression have been made, yet challenges remain. Traditional imaging techniques like MRI, CT, and mammography are limited by the need for professional curation, which is time-consuming. Genetic changes associated with cancer could serve as diagnostic, prognostic, and predictive biomarkers, but their translation into clinical practice is hindered by variations in metastasis, treatment responses, and resistance. New therapeutic strategies, while efficient, face issues due to cancer heterogeneity. Artificial intelligence (AI) offers solutions to these challenges, with extensive applications in drug development, cancer prediction, diagnosis, and the analysis of next-generation sequencing data. AI algorithms can identify genetic mutations or signatures for early cancer detection and targeted therapies. However, developing and implementing accurate AI models in clinical settings is challenging due to data heterogeneity, biases, and privacy concerns. Despite these, AI has demonstrated improved clinical decision-making.
Artificial intelligence, a collection of methods and techniques, has become increasingly important in cancer research, with various AI methods being detailed in this review, including their advantages and limitations. The review provides an overview of the usage of these methods over the past decade, as well as guidelines on incorporating AI models into clinical settings and the potential of pre-trained language models in personalizing cancer care strategies. AI methods discussed include machine learning (ML), which encompasses unsupervised and supervised learning. Supervised learning, which includes regression and classification, is widely used in cancer research. Traditional ML models like Bayesian networks, support vector machines, and random forests continuously incorporate data to produce outcomes. Deep learning, a subset of ML, uses multiple hidden layers to identify complex patterns in data. Natural language processing (NLP), another AI algorithm, targets narrative texts to extract useful information for decision-making.
AI models in cancer research utilize multi-omics and clinical information from various sources, with classification being the most common task. These models are validated and assessed using receiver operating characteristic analysis, which computes area under the curve (AUC), sensitivity, specificity, and precision. AI methods have been developed to handle large volumes of data, requiring increased cloud computing and storage power. The review also discusses the application of AI in drug development, where models predict drug responses using multi-omics data. Additionally, AI has been used to extract information from electronic health records, addressing the challenge of analyzing messy data.
Despite the progress, there are limitations to AI applications in cancer research. Choosing the appropriate algorithm is complex and depends on data type and complexity. Integrating AI into clinical settings requires detailed application explanations and transparency of algorithms. Monitoring the quality of AI tools for robust performance is crucial. The review emphasizes the need for further transparency and guidance on software scrutiny, cost-effectiveness, retraining of data sets, and conditions required for using AI systems.
In conclusion, AI has significantly impacted cancer research, and addressing challenges and validating AI-generated results can lead the future of oncology research. The review highlights the progress of AI methods in cancer-related applications and the potential of explainable AI, personalized medicine, and non-invasive AI tools for early cancer detection. As AI continues to evolve, it holds great potential in revolutionizing cancer detection and improving patient outcomes.
DOI: 10.1007/s11684-024-1085-3
