Leveraging Big Data to Enhance AI in Cancer Detection and Treatment
Integrating AI into the healthcare decision making process is helping to revolutionize the field and lead to more accurate and consistent treatment decisions due to its virtually limitless ability to identify patterns too complex for humans to see.
The field of oncology generates enormous data sets, from unstructured clinical histories to imaging and genomic sequencing data, at various stages of the patient journey. AI can “intelligently” analyze large-scale data batches at faster speeds than traditional methods, which is critical for training the machine learning algorithms that are foundational for advanced cancer testing and monitoring tools. AI also has tremendous inherent pattern recognition capabilities for efficiently modeling data set complexities. This is important because it enables deeper, multi-layered understandings of the impact of nuanced molecular signatures in cancer genomics and tumor microenvironments. Discovering a pattern between genes only found in a certain subset of cancer cases or cancer progression patterns can lead to a more tailored, patient-specific approach to treatment.
What is the ultimate goal? AI-powered cancer tests that support clinical decision-making for doctors and their patients at every step of the cancer journey – from screening and detection, to identifying the right treatment, and for monitoring patients’ response to interventions and predicting recurrence.
Data Quality and Quantity: The Key to AI Success
Ultimately, an AI algorithm will only be as good as the quality of data that trains it. Poor, incomplete or improperly labeled data can hamstring AI’s ability to find the best patterns (garbage in, garbage out). This is especially true for cancer care, where predictive modeling relies on impeccable precision – one gene modification out of thousands, for example, could signal tumor development and inform early detection. Ensuring that high level of quality is time-consuming and costly but leads to better data, which results in optimal testing accuracy. However, developing a useful goldmine of data comes with significant challenges. For one, collecting large-scale genomic and molecular data, which can involve millions of data points, is a complex task. It begins with having the highest quality assays that measure these characteristics of cancer with impeccable precision and resolution. The molecular data collected must also be as diverse in geography and patient representation as possible to expand the predictive capacity of the training models. It also benefits from building long-term multi-disciplinary collaborations and partnerships that can help gather and process raw data for analysis. Finally, codifying strict ethics standards in data handling is of paramount importance when it comes to healthcare information and adhering to strict patient privacy regulations, which can sometimes present a challenge in data collection.
An abundance of accurate, detailed data will not only result in testing capabilities that can find patterns quickly and empower physicians with the best opportunity to address the unmet needs for their patients but will also improve and advance every aspect of clinical research, especially the urgent search for better medicines and biomarkers for cancer.
AI Is Already Showing Promise in Cancer Care and Treatment
More effective ways to train AI are already being implemented. My colleagues and I are training algorithms from a comprehensive array of data, including imaging results, biopsy tissue data, multiple forms of genomic sequencing, and protein biomarkers, among other analyses – all of which add up to massive quantities of training data. Our ability to generate data on the scale of quadrillions rather than billions has allowed us to build some of the first truly accurate predictive analytics in clinical use, such as tumor identification for advanced cancers of unknown primary origin or predictive chemotherapy treatment pathways involving subtle genetic variations.
At Caris Life Sciences, we’ve proven that extensive validation and testing of algorithms are necessary, with comparisons to real-world evidence playing a key role. For example, our algorithms trained to detect specific cancers benefit from validation against laboratory histology data, while AI predictions for treatment regimens can be cross compared with real-world clinical survival outcomes.
Given the rapid advancements in cancer research, experience suggests that continuous learning and algorithm refinement is an integral part of a successful AI strategy. As new treatments are developed and our understanding of the biological pathways driving cancer evolves, updating models with the most up-to-date information offers deeper insights and enhances detection sensitivity.
This ongoing learning process highlights the importance of broad collaboration between AI developers and the clinical and research communities. We’ve found that developing new tools to analyze data more rapidly and with greater sensitivity, coupled with feedback from oncologists, is essential. Bottom-line: the true measure of an AI algorithm’s success is how accurately it equips oncologists with reliable, predictive insights they need and how adaptable the AI strategy is to ever-changing treatment paradigms.
Real-World Applications of AI Are Already Increasing Survival Rates and Improving Cancer Management
Advances in data scale and quality have already had measurable impacts by expanding the physician decision-making toolkit, which has had real-world positive results on patient care and survival outcomes. The first clinically validated AI tool for navigating chemotherapy treatment choices for a difficult-to-treat metastatic cancer can potentially extend patient survival by 17.5 months, compared to standard treatment decisions made without predictive algorithms1. A different AI tool can predict with over 94% accuracy the tumor of origin for dozens of metastatic cancers2 – which is critical to creating an effective treatment plan. AI algorithms are also predicting how well a tumor will respond to immunotherapy based on each person’s unique tumor genetics. In each of these cases, AI toolkits empower clinical decision-making that improves patient outcomes compared with current standards of care.
Expect An AI Revolution in Cancer
AI is already changing how early we can detect cancer and how we treat it along the way. Cancer management will soon have physicians working side-by-side with integrated AI in real time to treat and monitor patients and stay one step ahead of cancer’s attempts to outwit medicines with mutations. In addition to ever-improving predictive models for detecting cancer earlier and providing more effective personalized treatment paradigms, physicians, researchers, and biotech companies are hard at work today to leverage data and AI analyses to drive new therapeutic discoveries and molecular biomarkers for tomorrow.
In the not-too-distant future, these once-impossible advances in AI will reach far beyond cancer care to all disease states, ending an era of uncertainty and making medicine more accurate, more personalized, and more effective.
