The future is now: AI and more effective cancer screening


Ben Hargreaves examines how AI cancer diagnosis is currently being implemented, and how it could evolve in the years ahead, including the potential for the exploration of digital biomarkers.

There have been discussions for a number of years over how the use of AI could revolutionise the healthcare industry. In recent years, there have been some concrete steps that have seen the technology move from buzzword to an actual focal point for industry, based on successes seen. The first AI-designed molecule, created by Insilico Medicine, entered trials at the beginning of 2022 and posted positive results from a Phase I trial at the beginning of 2023. The pharma industry is also looking at ways for AI to match patients to trials, as well as improve the efficiency of the manufacturing process.

Another way in which this technology is already having an impact is by assisting in scans of patients for cancer. Basic forms of AI have actually been used in some types of cancer scans for more than 20 years, but the technology has advanced rapidly in recent years, providing the capacity to spot cancer far earlier than had previously been possible. Only two years ago, research was published that found AI was able to outperform radiologists in mammographic screening. The question is how this approach can be broadly applied in cancer, and where it could lead in the future.

The early benefits

Reaching out to Paige, the first company to receive US Food and Drug Administration (FDA) approval for a clinical AI application in digital pathology, pharmaphorum wanted to better understand what the advantages are when using AI for cancer diagnosis. David Klimstra, chief medical officer of the company, explained: 

“Diagnostic AI can improve a number of aspects of pathological diagnosis. By quickly drawing the pathologist’s attention to suspicious findings, it makes the review more efficient, allowing the pathologist to focus their time on the most critical aspects of diagnosing cancer, rather than the time-consuming and exhausting process of screening numerous uninformative areas to identify the key finding, which is often the proverbial ‘needle-in-a-haystack’.”

The benefits of this approach do not end at freeing up time for the pathologist, as the AI can also allow for triaging of cases. This means that those patients who have the most important findings can be seen first, and follow-up studies can be ordered immediately, avoiding potential delays. Perhaps more importantly, Klimstra stated that AI is objective and therefore can help pathologists be more consistent, reducing the ‘interobserver’ variability that can impact subject diagnoses, such as cancer grading.

For more detail, pharmaphorum spoke to AstraZeneca, just one big pharma company that has invested into AI capabilities, including recently backing an AI-based lung cancer diagnosis pilot in the UK, together with This project involved working alongside radiologists and applying AI to scans, unlike Paige’s work with digital pathology – though, AZ is also active in the area. Ti Hwei How, VP, International Oncology at AZ, explained what is brought to the industry by the AI technology behind the collaboration: 

“Applying AI to cancer can lead to earlier diagnosis by helping to increase the detection of potentially actionable nodules and abnormalities more efficiently. This is important because early diagnosis gives patients increased treatment options and a better prognosis, as well as an opportunity to live fuller lives. Additionally, early diagnosis can decrease costs for health systems by enabling treatment that is generally more effective, less complex, and not as expensive.”

Searching for digital biomarkers

The use of AI for diagnosis is only just becoming utilised as a part of cancer care, but there are far larger ideas for how the potential of this area could unfold. David Klimstra outlined a future where the use of algorithms permeates routine practice, providing all cancer patients with the additional reassurance provided by a confirmatory review through AI technology. He also added that Paige expects to be able to have applications running across the spectrum of neoplastic diseases. Beyond these aspects, Klimstra believes the technology can go beyond reviewing scans: 

“The most exciting applications will involve digital biomarkers, where AI is used to extract additional information from the routine pathology slides that is not apparent based on current approaches. These digital biomarkers can replicate existing immunohistochemical or molecular biomarkers, in which case they can be used for screening cases for definitive testing, since digital biomarkers are fast, relatively inexpensive, and not tissue-consuming (they run on the same slides used for routine diagnosis).

“Additionally, digital biomarkers can provide novel information that is not available based on any current testing modality. Prediction of patient outcome, response to targeted treatment, or resistance to certain therapies will significantly improve the management of cancer patients, using only the routine pathology slides. These novel biomarkers are trained based on the clinical outcome that would be useful to predict, so they can supersede any existing tests.”

Improving global health

One area that perhaps does not come to mind with an emerging technology of the degree of sophistication of AI is the potential to reduce inequality of health outcomes. AZ has partnered with to also provide early stage diagnosis of lung cancer in Latin America, Asia, and Middle East and Africa regions. Ti Hwei How outlined the purpose:

“Our goal is to extend our reach to underserved and hard-to-reach populations to ensure more equitable access to cancer services and to transform what it means to be diagnosed with the disease. AI has allowed us to make progress in delivering equitable access to screening by enabling scalable technology in low- and middle-income countries where it was previously not possible.”

How explained that the AI is used to analyse X-rays for early-stage lung cancers, with the initiative integrating the technology into routine care and existing screening programs. In this manner, AI can improve detection of early stage cancer quickly and at scale. In terms of statistics, How stated that the company has already processed almost 128,000 scans for lung cancer risk, across 25 markets. The technology had identified lung nodules with high malignancy risk in 1.6% scans, discovering a risk to health which may not have been identified without the AI technology. 

Though 1.6% may seem a small percentage, this is the benefit of AI – wherein it can improve the odds of catching a warning sign that could be missed otherwise. Particularly when scaled out to encompass broader populations, across the world, such a small percentage could mean the difference between life and death for a huge number of individuals. When coupled with the potential to identify digital biomarkers, and help tailor treatments to patients, AI could soon become an essential tool to the pharma industry and assist in improving overall global health.