Artificial intelligence (AI) has emerged as the “connective tissue” that will facilitate oncology practice and deliver enhanced care to patients, according to Arturo Loaiza-Bonilla, MD, MSEd, FACP.

Loaiza-Bonilla, systemwide chief of Hematology and Oncology at Saint Luke’s University Health Network, spoke with CancerNetwork® about his insights regarding recent advancements for AI in oncology and provided a recap of some of the key developments with these technologies in 2025.

Loaiza-Bonilla began by discussing the application of AI in radiology scans, particularly for mammograms and early detection of cancer in lung nodules, as well as in predictive models for pancreatic cancer detection, which will enhance screening speed and efficiency and eliminate the need for double review on scans. Additionally, he cited ventures, such as ai.RECIST by Friends of Cancer Research, which aim to embed these algorithms into measurements for lesions.¹

He further touched upon AI’s implementation into cardio-oncology through foundational models, which will help detect cardiovascular abnormalities in patients undergoing treatment for cancers. Additionally, Loaiza-Bonilla further discussed advances in breast cancer screening, such as population-level validation, recurrence prediction models, and radiomics to determine response to therapy. Moreover, he highlighted an ability for AI to democratize pathology by providing biomarkers, even in the absence of a pathologist.

He concluded by providing his thoughts regarding AI’s evolving role in oncology, as well as a need to ensure that tools for patient and clinician use are properly regulated and vetted for accurate use and efficiency to better deliver guideline-concordant care.

CancerNetwork: How is AI being used to supplement scans conducted by radiologists?

Radiologists now, particularly in oncology, are mostly focused on 2 major trends. Mammography is one of them; [we] try to optimize times because there are many patients who require screening and [it’s a] significant amount of [time and] work. We are now using AI to assist with that, to decrease the amount of double review on those mammograms, and it’s also being run on clinical trials. We have data from Europe, but now there’s a number of efforts in the US as well, because our population is different in terms of breast density and beyond. I’m excited to see the results of those algorithms as they get deployed.

Other [trends] that we are looking [at] closely are in finding lung nodules and detecting them in real time so we can find lung cancer early through screening, even from X-rays found in the emergency department. There are some efforts doing [that]. More advancement is on the use of RECIST criteria for clinical trials. There’s an effort from Friends of Cancer Research where they are trying to do ai.RECIST criteria, where we just embed the algorithms and have been able to measure the lesions in a more effective and scalable way. Radiologists are a key part of that.

Lastly, [AI is being used in] some predictive models. There are some efforts, like the PANORAMA trial results, where they show that … CT scans can detect pancreatic cancer early.² This is one of the diseases that we want to find them early, before they really become metastatic, as most of them are either locally advanced/resectable or metastatic. Very promising on the radiology side.

What I hope for is to start using those models in real practice. As of now, many of the models that we use in radiology have been localized to those couple use cases and for the emergency department, such as finding fractures, but now, with oncology being one of those very high stakes diseases, we have an opportunity here.

How might the inclusion of EKG AI models help facilitate cardio-oncology processes and improve outcomes for patients with cancer?

These topics are about foundational models. Foundational models mean very large data sets that include hundreds or hundreds of thousands, sometimes millions, of tests, all labeled and being built for [generalizability] to a specific use case. In this case, it could be the detection of arrhythmias or risk factors such as Qt prolongation in patients taking [tyrosine kinase inhibitors] or [adverse] effects that are emerging, particularly for patients who may have predispositions such as cardiovascular disease.

The use of EKGs in foundational models is becoming a new trend. There are several companies doing so at this moment, trying to show the FDA the validity of those tests and those foundational models. What I anticipate is, because these EKGs are so easy to use now, they’re ubiquitous, we hopefully can help to do some clearances for either potential surgeries the patients may have, or if they are on treatment, predicting the risk of having complications from the drugs that we use. It’s a combination of the EKG models plus other algorithms that are more data-driven in terms of integration with the [electronic medical record (EMR)], which are going to be helpful to really move the needle. More to follow on that; we are going to probably be hearing more in the next year or so.

What advances in breast cancer screening have been made in 2025, and what are the next steps for research and development in this area?

In breast cancer, the new steps are, first, validation at the population level. Each country and each region is doing its own clinical trials. I’m currently running for early detection with mammograms and others. We also have seen that the implementation of AI into the genomic testing, or the biomarker testing, done on our patients is becoming relevant.

There were some updates from [the San Antonio Breast Cancer Symposium] showing how a potential transformer-based architecture can predict potential recurrence in patients. It can optimize those recurrent response scores that we are seeing from different vendors in a more meaningful way, and find those niche patients that need either more screening or early detection of other things, such as germline alterations or risk factors, when we use any combination with other data sets.

In this case, for example, we can use radiomics. We know that with radiomics, we can, just by looking at a CT scan, understand if the patient is going to respond to, for example, endocrine therapy. That’s something that, for us, is mind blowing, because we did not have that information before, but now, using the data sets for many clinical trials that have been done before––the [phase 3 TAILORx trial (NCT00310180)] and others––that has been a major validation point for many of those models that are being deployed in the clinic, or at least in the clinical practice for further validation.³

How might AI help democratize pathology, particularly among resource-constrained regions, and how do foundational models lead to the development of tools with better generalizability?

We are, at this moment, very short globally, not only in low- and middle-income countries, but in counties in the United States that do not have pathologists readily available to review slides and give us support. One of the things that we’ve seen now is with these big foundational models of millions of slides being taken, whole slide imaging pictures, we can leverage those models to look at just the H&E stain. Sometimes even without the H&E stain, just simply the slide itself can give us biomarkers. We may have the ability to get immediate readouts, but we have an algorithm deployable that is just sent by cloud and [can be] downloaded into a local computer that everyone usually has. Then it can really solve a lot of problems in terms of first screening for, for example, malignancy. Someone needs to be seen right away for additional treatment options. And we know about these deserts of cancer care where we don’t have an oncologist or pathologist within 100 miles, which is crazy and even happens in the United States.

If we can use these tools at scale, where we just simply have the slide that anyone has…you simply take the picture and deploy and say, “Hey, this may have an EGFR mutation or this sample may be [estrogen receptor]/[progesterone receptor]-positive or have PL1 positivity.” It can be consequential in the use of the pathology. I believe that multidisciplinary efforts need to happen across [practices] to implement this, but the promise is tremendous.

What other advances in AI’s application to oncology do you believe will have the potential to transform practice for patients?

2026 is going to be a key year because we are seeing a signal of what I call the industrialization of AI in oncology. There are no longer services that are infrastructure embedded; even pharmaceutical companies and health systems are investing in licensing the foundational models, and training them on their own proprietary data sets and putting that into the drug discovery, traditional workflows, and Agentic AI for the final determination of steps. The idea here is to make this full loop closer together from, drug discovery to getting the drug in front of the patient. That’s the goal post now.

We are seeing discovery cycles being compressed from years to now, sometimes months, and in a few cases, weeks. That is how we are going to see an improvement. From the clinical aspect, AI is becoming more like a connective tissue in the oncology continuum. It’s not about, “Should I adopt it?” It’s more, “How do I adapt my workflow to the advent of AI to the benefit of my practice and my patients?” We have seen Agentic AI supporting tumor boards, clinical trial matching, and we do pathway adherence, care coordination, and embedding with guidelines in ASCO and NCCN.

It’s not replacing physicians; it’s removing friction points that are happening in that continuum. The most meaningful deployments, I believe they are not going to be flashy or press releases. It is going to be if we can reduce documentation burden, if we can accelerate triage, if we can get access to trials faster and help teams execute, that is going to be the real value that we’re going to see in deployments of these tools.

Is there anything else you would like to discuss that we might not have already covered?

Something important is that we are seeing the integration of AI also for patients. We have chatGPT Health, we have Claude, and many other patient-centric approaches, which are important because a lot of patients are seeking care through those outputs. The opportunities are huge, but they need to be aligned with evidence, with safety, with workflows, and of course, we want to avoid any misinformation.

That’s why I feel there has to be clinical involvement, governance, and some guardrails, at least with specialty-driven pathways. Hopefully, that translates into what I feel are the best use cases, which are speed, access, and execution. Did it bring therapies to patients faster, or were they very informed? Can we get access to clinical trials and guideline-concordant care faster? Did my team and [I] spend more time practicing medicine and less time fighting systems? If AI helped us do those things, they have earned their place in oncology. That’s how it’s going to come from a peripheral technology to an interesting foundational component of our workflows.

Source: Taking Stock of AI Advancements in Oncology: Present and Future | CancerNetwork

References

1. Ai.RECIST Project: Artificial Intelligence-Enabled Response Evaluation Criteria in Solid Tumors Project. Friends of Cancer Research. Accessed January 28, 2026. https://tinyurl.com/rx26459z
2. Alves N, Schuurmans M, Rutkowski D, et al. Artificial intelligence and radiologists in pancreatic cancer detection using standard of care CT scans (PANORAMA): an international, paired, non-inferiority, confirmatory, observational study. Lancet Oncol. 2026;21(1):116-124. doi:10.1016/S1470-2045(25)00567-4
3. Krishnamurthy S, Feng C, Muhammad H, et al. Artificial intelligence predicts OncotypeDX recurrence scores directly from H&E-stained whole slide images of ER+/HER2- node-negative breast cancer surgical sections. Presented at: 2025 San Antonio Breast Cancer Symposium; December 9-12, 2025; San Antonio, TX. PS3-06-02