top of page

The Vital Role of AI In Diagnosing Cancer


Siddartha Mukherjee famously labeled cancer "the emperor of all maladies, the king of all terrors." Responsible for countless deaths, pushing trauma and disruption into the lives of millions of families, and with treatment being extremely expensive, in many ways, the diseases that are cancer still elude us. In many cases, this comes down to the fact that we don't detect cancer early enough to enable effective treatment or know how to tailor that treatment to the individual case before us.

As Mukherjee explains, "there are victories and losses, campaigns upon campaigns… survival and resilience." Often the individual's cancer experience is dubbed a 'battle,' but the war is that of healthcare versus this nefarious malady rather than the individual plagued by the disease. Mukherjee likens our ongoing quest to fight cancer to a military one. It is the oncological arms race; hitherto, we haven't had good enough weapons.

The next frontier is artificial intelligence (AI). AI can analyze immense amounts of data to reveal patterns, matches, and insight. An excellent example of developing AI as a weapon in the battle against cancer is the Galleri Test from GRAIL which uses machine learning to analyze a single blood draw to detect differences between methylation patterns in DNA from cancer and non-cancer cells.

Here I look at AI's crucial role in cancer diagnosis, looking at existing developments and the space in which AI can transform our approach to cancer diagnosis and treatment in ways we could merely imagine a few decades ago.

The Power of Numbers In Imaging

Already, AI has made headway by utilizing reams of data in the form of images for cancer detection. With enough data and appropriate algorithms, AI can spot anomalies in tissue sample images. Indeed, AI can increasingly spot changes and abnormalities more reliably and earlier than the trained but naked eye. Algorithms must feed on thousands of images.

We have witnessed an example of AI’s benefit with research from Tulane University. Researchers trained their algorithm on more than 13,000 images of colorectal cancer. Subsequent attempts to use this learning to detect new cases of colorectal cancer succeeded with an incredible 98% accuracy. Most skilled oncologists would fall considerably short of this success rate!

As such, AI algorithms can increase accuracy in detection. Benefits extend further, with oncologists freed from the time-consuming task of image review, allowing them time to better attend to treating patients. Patients also benefit from a reduced need for invasive biopsies and shorter anxiety-laden waiting times.

Another excellent example of the use of AI in cancer diagnosis comes from the Laura and Isaac Perlmutter Cancer Center with the Department of Radiology at NYU Langone Health. Here, researchers using AI achieved a 37% improvement in accuracy alongside a 27% reduction in tissue sampling.

The use of AI in cancer diagnosis doesn’t come without pitfalls, and it will take more time before we see broad uptake and use of these technologies. However, this is an up-and-coming area with many motivating factors.

Learning about risk

It is one thing to detect and diagnose cancer, but it's a step further to analyse risk and fully understand it. This is because multiple factors come into play to determine an individual's cancer risk, including genetics and lifestyle.

One such factor is prognostic markers. These biological markers can reveal higher risks for developing different types of cancer and the patient's outcomes. An example is tissue density evidenced in mammograms, where increased density is associated with increased cancer risk. Another example is the existence or activation of specific genes linked to some cancers, such as TP53, WT1, Ki67, Topo-II, BRCA1 and BRAC2 for ovarian cancer.

Artificial intelligence has gained ground by analysing such risk factors. Proof of this in practice comes from researchers at the Universities of Iowa and Wisconsin who have used artificial neural networks (ANN) to improve the ability to predict disease recurrence and thus better understand prognosis.

Again, machine learning is more effective than the naked eye here, picking up previously unidentified novel markers. Furthermore, through the full use of AI, whereby the tech can create its own predictive models, other valuable markers were revealed that weren't even on the researchers' radar.

Perhaps one of the most widely known examples of AI in cancer diagnosis is the 2021 University of Hawaii research using machine learning with mammogram imaging. An algorithm trained on 6,369 women's images proved capable of predicting the later emergence of interval cancers with notable improvements compared to traditional clinical risk factors such as tissue density.

Dr John Shepherd, a co-author of the research, stated, "The results showed that the extra signal we're getting with AI provides a better risk estimate for screening-detected cancer. It helped us accomplish our goal of classifying women into low risk or high risk of screening-detected breast cancer."

Such knowledge should, ultimately, give oncologists more and better tools which can be used for both early detection and assessing future risk.

Drilling down and getting personal

Cancer is a vast umbrella term encompassing hundreds of unique diseases and even more presentations and outcomes. For a generic term, it's painfully personal. A unique mix of genetic, environmental and disease characteristics characterises each case. This makes for a complex picture for detection, diagnosis and treatment.

AI also offers promise here as the ability to drill down into specific cancer sub-classifications and accurately predict disease progression, and the individual's treatment receptivity would transform care and outcomes. Already multiple exciting cancer products led by Microsoft reveal promise in this respect:

● InnerEye combines machine learning and natural language processing helping oncologists and radiologists better understand how tumours progress.

● BC Cancer and Microsoft Canada’s “Single Cell Genomics” allows physicians to see the genomes of individual cancer cells. This enables tailored and directed treatments. It also helps predict how a patient may respond to chemotherapy.

● Microsoft’s cloud-based tool, Bio Model Analyzer, models cells so we can understand how they interact, communicate and connect, allowing earlier cancer detection. It also enables a better understanding of cancer treatment by using models which predict therapy efficacy and resistance.

● “Project Hanover” is a collaboration between Microsoft and AstraZeneca which processes fragmented information related to drug interactions and resistance in leukaemia patients, identifying relevant data for oncologists to develop tailored treatment plans.

However, the most exciting early cancer diagnosis is from Grail. With a single blood draw they can predict over 50 types of cancer in people with no symptoms and 45 of these currently have no other diagnostic screening available. Their clinical trials have been conducted on the largest population study of its kind and the results have had high predictive accuracy. When the cancer signal is detected in their Galleri test, it localizes the cancer signal with high accuracy, helping inform the next steps to the diagnosis of the type of cancer detected. In fact, companies such as Novartis (a pharmaceutical company with a strong oncology portfolio) have offered the Galleri test to all their staff.

These tools and others like them radically transform our understanding of cancer at the micro level, leading to impressive early detection and stronger patient outcomes.


Artificial Intelligence has immense potential to transform our approach to the “emperor of all maladies.” It allows us to enter a new realm in the war on cancer by approaching from another angle – detection and diagnosis and tailored treatment planning. Already, we can see AI in action, making notable inroads with earlier detection, better diagnosis and tailored treatments.

It's exciting that AI will provide more weapons in our arsenal as this field develops. Perhaps AI will allow us to banish cancer from the battlefield altogether in the coming decade.

Recent Posts

See All


bottom of page