Star Trek promised us a medical tricorder that could wave over a patient and spit out answers like a smug little oracle. This paper argues we may be building the scrappier, real-world version out of two very ordinary things: pathology slides and blood tests. Not quite starship sickbay, but close enough to make an old sea dog raise an eyebrow and check the compass twice.[1]

The basic problem in precision oncology is not that we lack fancy maps. It is that the best maps often cost a fortune, take too long to print, and never reach half the ports that need them. A lot of modern cancer treatment depends on sequencing-based biomarkers, which can help doctors match patients to therapies. Fine idea. But in much of the world, that route is slow, expensive, or simply not available. Hoang and colleagues make the case that AI could chart a cheaper course by squeezing useful predictions out of routine tissue slides and standard blood work instead.[1]

Old Charts, New Seas

Histopathology is the business of turning a tissue sample into a stained slide and asking a pathologist, "What on earth am I looking at?" Digital pathology takes that slide, scans it, and lets algorithms join the watch crew. Recent reviews show these models can help with diagnosis, grading, mutation prediction, treatment planning, and survival estimation.[2,3] A 2024 meta-analysis covering more than 152,000 whole-slide images found high average sensitivity and specificity across many disease areas, which is impressive even if the literature still has enough bias and reporting gaps to make a statistician reach for the lifeboat.[4]

Blood-based biomarkers are the other half of the story. Think of them as messages in bottles floating through the bloodstream: proteins, circulating tumor DNA, inflammatory markers, and other molecular scraps that whisper what the tumor is up to. Liquid biopsy already matters in oncology because it is less invasive and easier to repeat over time than cutting out tissue every few weeks.[5] Add AI, and those bottles stop being random flotsam and start looking more like a navigational system.

That is the wager in this paper. Instead of asking every patient to pass through high-cost genomic testing, perhaps we can use routine pathology and blood data as a faster first-pass signal for prognosis and treatment response. In plain English: less waiting, less cost, more reach. If that works reliably, it is a very big deal.

Why This Matters Outside the Fancy Harbors

The most interesting part of this paper is not the algorithmic swagger. It is the access argument. Precision oncology has had a bad habit of acting like a luxury cruise. Beautiful decks, excellent instrumentation, and a ticket price that quietly excludes most of humanity. AI on routine slides and blood tests offers a different vessel: less glamorous, more practical, and maybe far better at reaching under-resourced clinics.

That fits the broader weather report in the field. Recent commentaries and reviews in 2024 and 2025 keep circling the same point: AI biomarkers are becoming more believable, multimodal models are pulling together text, images, and molecular data, and pathology is turning into a major engine room for treatment decisions.[3,6] Conference reporting from ESMO and ASCO in 2025 also showed growing interest in AI-derived biomarkers for treatment response and subtype scoring, which suggests the fleet is moving from demo harbor toward real clinical waters, albeit slowly and with much swearing below deck.[7,8]

No Magic Spyglass, Though

Before anybody starts yelling that the machine can "see cancer secrets invisible to humans," let us reef the sails a bit. The technology is promising, but medicine is where overconfident software goes to get humbled.

Models trained on slides from one hospital can drift badly when they meet data from another. Staining varies. Scanners vary. Populations vary. Reporting quality varies. Sometimes the model is learning tumor biology; sometimes it is learning that Hospital A likes a particular shade of pink. Recent reviews have been blunt about this: external validation, generalizability, and workflow integration are still rough seas.[2,4,6]

Blood-based models face similar trouble. Biomarkers can be noisy, patient populations messy, and the temptation to overfit enormous. I have seen many a model founder on the rocks of "look, the AUC is amazing" right before someone asks whether it works on patients outside the original dataset. That question is the iceberg. Ask it early.

The Course Worth Following

Still, the direction here feels right. If cancer care is going to become more precise, it also has to become more available. A biomarker that works only in wealthy hospitals with deep sequencing budgets is not much of a lighthouse for the rest of the world. But a tool that can extract useful guidance from routine slides and blood tests? That starts to look like a chart a lot more clinics could actually use.[1]

So no, this is not the sci-fi tricorder yet. It is more like a sturdy navigation kit assembled from microscopes, blood tubes, and a lot of math done by silicon deckhands who never sleep. Messy? Absolutely. Worth watching? Aye.

References

1. Hoang DT, Chang TG, Ferrone CR, Ronai ZA, Ruppin E. AI-Driven Pathology and Blood-Based Biomarkers: A Golden Opportunity to Democratize Precision Oncology. Cancer Discovery. 2026. DOI: 10.1158/2159-8290.CD-25-2065. PubMed: 42008780

2. Aggarwal A, Bharadwaj S, Corredor G, et al. Artificial intelligence in digital pathology - time for a reality check. Nature Reviews Clinical Oncology. 2025;22:283-291. DOI: 10.1038/s41571-025-00991-6

3. Bahadir CD, Omar M, Rosenthal J, et al. Artificial intelligence applications in histopathology. Nature Reviews Electrical Engineering. 2024;1:93-108. DOI: 10.1038/s44287-023-00012-7

4. McGenity C, Clarke EL, Jennings C, et al. Artificial intelligence in digital pathology: a systematic review and meta-analysis of diagnostic test accuracy. npj Digital Medicine. 2024;7:114. DOI: 10.1038/s41746-024-01106-8

5. Liquid biopsy. Wikipedia. Accessed April 25, 2026. https://en.wikipedia.org/wiki/Liquid_biopsy

6. Truhn D, Eckardt JN, Ferber D, et al. Large language models and multimodal foundation models for precision oncology. npj Precision Oncology. 2024;8:72. DOI: 10.1038/s41698-024-00573-2

7. ESMO Daily Reporter. New AI-based biomarkers offer insight into treatment response. Published October 2025. https://dailyreporter.esmo.org/esmo-congress-2025/ai-digital-oncology/new-artificial-intelligence-based-biomarkers-offer-early-insight-into-treatment-response

8. The Pathologist. AI in Pathology: The Six Pillars of Progress. Published August 2025. https://www.thepathologist.com/issues/2025/articles/aug/ai-in-pathology-the-six-pillars-of-progress/

Disclaimer: This blog post is a simplified summary of published research for educational purposes. The accompanying illustration is artistic and does not depict actual model architectures, data, or experimental results. Always refer to the original paper for technical details.