Why Is Cancer So Hard to Cure? Can AI Help Detect It Earlier?

Cancer is one of the most studied diseases in medicine, yet we still talk about "the cure" as if it were a single problem waiting for a single answer. In our short video explainer, Why Is Cancer So Hard to Cure? Can AI Help?, we look at why that framing breaks down, how doctors find cancer today, and where blood-based DNA tests and artificial intelligence may help detect it earlier. This article expands on the video with more detail and sources, so you can watch first and read on for the depth.

Cancer is not one disease

The US National Cancer Institute (NCI) counts more than 100 types of cancer. What they share is a root cause: cancer is a genetic disease, caused by changes to the genes that control how our cells grow and divide. Those changes can come from errors made when cells copy their DNA, from damage caused by substances in the environment, or from variants inherited from a parent. Because the affected gene, tissue and cell type differ from one cancer to the next, so do the way the tumor behaves and the treatments that work against it. A drug that works well against one cancer can be useless against another.

Tumor diversity: why one person's cancer is not another's

Each person's cancer has its own combination of genetic changes, and as a tumor grows it keeps accumulating new ones. NCI notes that even within a single tumor, different cells can carry different genetic changes. This is called tumor heterogeneity, and it shows up at several levels:

• Between cancer types: lung, breast, colon and blood cancers are driven by different genetic changes and behave differently
• Between patients: two people with the same type of cancer can carry different mutations
• Within one tumor: different groups of cells can respond differently to the same drug
• Over time: a tumor keeps changing, so what was true at diagnosis may not be true after months of treatment

Why treatment is so difficult

Diversity is only part of the problem. Three challenges come up again and again in cancer care:

• Resistance: treatment can eliminate sensitive cells while resistant cells survive and regrow, so a tumor that shrinks at first can return
• Spread (metastasis): cancer that has spread to other organs is much harder to remove or control, and NCI notes that most people who die of cancer die of metastatic disease
• Late detection: cancers found while still small and localized are generally easier to treat than those found after they have spread, but many cancers cause few symptoms early on

How cancer is diagnosed today: scans and biopsies

When a doctor suspects cancer, imaging such as MRI or CT can show where an abnormal area is and how large it is, but imaging alone usually cannot say what the tissue actually is. The standard way to confirm a diagnosis is a biopsy: a sample of the suspicious tissue is removed and examined by a pathologist, and increasingly its DNA is analysed as well. Biopsies are powerful, but they are invasive, they sample only one part of a tumor that may be diverse, and they can be difficult or risky to repeat. That is why researchers are looking for less invasive ways to get the same information.

Liquid biopsy: reading tumor DNA in a blood sample

Tumors shed small fragments of DNA into the bloodstream. A liquid biopsy looks for tumor cells or those DNA fragments, called circulating tumor DNA (ctDNA), in a blood sample. Because the DNA carries the genetic changes of the tumor it came from, sequencing it can in principle reveal what is happening in the body without surgery. Today, liquid biopsies are used mainly in specific clinical settings, such as guiding treatment choices and monitoring cancer in people who already have a diagnosis. Using them to find cancer in healthy people is a much harder problem and is still being tested. Early detection is difficult for several reasons:

• Early tumors are small and shed very little DNA, so the signal can be faint
• Healthy blood also carries DNA from normal cells, including harmless age-related changes, which creates background noise
• A false alarm can cause anxiety and unnecessary tests, while a missed cancer can create false reassurance
• Any screening test has to be proven in large studies to actually help people, not just to find more abnormalities

Where AI fits in

Sequencing a blood sample produces enormous amounts of data, and the cancer signal inside it can be subtle. This is where machine-learning models are being tested: they can learn patterns in DNA, such as mutations or chemical marks called methylation, that separate a cancer signal from background noise, and they can help estimate where in the body a signal might come from. AI is also being studied in other parts of cancer care:

• Imaging: in the MASAI randomized trial of more than 100,000 women in Sweden, AI-supported mammography screening detected a higher share of cancers at screening than standard double reading by radiologists (81% versus 74% of cancers diagnosed), with a similar false-positive rate (1.5% versus 1.4%) and 12% fewer cancers appearing between screening rounds
• Digital pathology: software that helps pathologists examine tissue slides
• Genomics: helping interpret which DNA changes in a tumor matter for treatment
• Research: helping prioritise drug targets and match patients to clinical trials

What still needs to improve

Our video ends on the same honest note: the technology is promising, but it is not ready to replace standard screening. The main hurdles are proving benefit in large trials, keeping false positives low, making sure tests work across different ages, ethnic groups and cancer types, and deciding what to do when a test finds something. Multi-cancer early detection blood tests are the clearest example. As of 22 September 2026, no such test has been approved by the US Food and Drug Administration (FDA); GRAIL's Galleri test is under FDA review, with an FDA advisory committee meeting scheduled for 23 September 2026. Until the evidence has been reviewed, these tests are not a substitute for recommended screening.

What this means for you today

Screening tests with proven benefit, such as mammography, colonoscopy or stool-based tests, cervical screening and, for some people, low-dose CT for lung cancer, remain the best-supported way to find certain cancers early. Which ones apply to you depends on your age, sex and family history, so ask your doctor.

• Follow your doctor's screening recommendations for your age and risk
• Tell your doctor about any family history of cancer, because it can change when screening starts
• Do not ignore symptoms that persist, such as a new lump, unexplained weight loss, bleeding where there should be none, or a cough or pain that does not go away
• Be cautious with direct-to-consumer cancer-detection tests, and discuss any result with a clinician

Understand your own lab results

If you have blood test results in front of you, our free interpreter explains each value in plain English against reference ranges and tells you when it is worth following up with a doctor. It is an educational tool: it cannot diagnose cancer or any other condition. Use the Interpret My Results button in the sidebar to try it.

What should we explain next?

Which topic should we cover next: cancer blood tests, immunotherapy or AI in cancer detection? Tell us in the comments on the video, and subscribe to Bioinfoxpert on YouTube for more explainers on AI, biology and the science shaping healthcare.