The new detection frontier
Galleri, Cancerguard and a fast-growing field of DNA, protein, RNA, cell, urine, breath and new imaging tests aim to detect cancer before symptoms.
Named tests to watch
These products use different signals, thresholds and study populations. A headline cancer count or sensitivity cannot compare them fairly.
GRAIL uses cell-free DNA methylation and machine learning to detect a cancer signal and predict its likely origin across more than 50 cancer types.
Exact Sciences combines methylation signals with selected proteins. It is the commercial successor to the CancerSEEK research program.
Guardant combines mutations, methylation and fragment patterns for average-risk colorectal screening. It is not a multi-cancer test.
FDA decisionCombines several tumour-associated proteins with age, sex and machine learning. It is lower-cost, but differs fundamentally from DNA-first tests.
Developer informationThese methylation and fragmentomics platforms are being evaluated across Asian populations. Transportability and outcome evidence remain unsettled.
Published studiesTrucheck assesses circulating tumour-associated cells; PanTum examines markers within activated immune cells before targeted imaging. Independent evidence is less mature.
Independent comparisonOn the horizon
Every technology is labelled by the strongest milestone actually reached, not the developer’s eventual ambition.
Midjourney Medical has scanned real volunteers in an IRB-reviewed body-composition study. A person is lowered into warm water as a ring of sensors traverses the body.
Private whole-body MRI screening is growing in visibility because it images extensive soft tissue without ionising radiation. It can find unsuspected cancers, but also benign cysts, nodules and anatomical variants.
Laser pulses generate ultrasound from light-absorbing tissue, revealing oxygenation and tumour-associated blood vessels at greater depth than optical imaging alone.
Tumours release tiny membrane packages containing DNA, RNA, proteins and lipids. They may preserve signals that are too scarce or fragile in free circulation.
Electronic noses and mass spectrometry seek patterns of volatile organic compounds produced by tumours, metabolism, microbes or inflammation.
The incidentaloma problem: whole-body imaging may reveal lesions that are real but harmless, indolent or impossible to classify immediately. A person may hear “we found something” and then face weeks of anxiety, repeat imaging, specialist review or biopsy before cancer is confirmed or excluded. Trials must count this diagnostic burden alongside cancers detected.
No “FDA registered” shortcut: establishment registration, device listing, breakthrough designation, laboratory certification and FDA clearance or approval mean different things. The exact indication matters.
Beyond a mutation test
Most use sequencing rather than simple PCR alone. Many combine weak biological signals with machine learning.
Looks for methylation patterns, fragments of circulating DNA, proteins or other signals across many cancers, often with a predicted tissue of origin.
Can profile mutations in known cancer, help select treatment and detect molecular residual disease. Its role in directing postoperative treatment is still being tested.
A machine-learning score combines circulating and exosomal microRNAs with CA 19-9. In an international observational biomarker study of 1,785 people, it detected 86.8% of stage I–II pancreatic cancers in the testing cohort.
The FDA approved Shield in 2024 for average-risk adults aged 45 and over. A positive test requires colonoscopy; it detects fewer precancerous lesions than colonoscopy-based pathways.
Urine cytology and selected biomarker tests can support bladder-cancer assessment. DNA, RNA, protein and metabolite tests for bladder, kidney and prostate cancer remain active research areas.
Proteins, DNA, RNA, microbes and metabolites are being evaluated, especially for oral and head-and-neck cancers. Standardized validation in real screening populations is still needed.
Mass spectrometry, sensor arrays and electronic noses seek cancer-related patterns in exhaled compounds, especially for lung and gastrointestinal malignancies.
SoftVue and QT Imaging scan a pendant breast in a warm water bath using rotating reflection and transmission ultrasound to build quantitative 3D images without compression or ionising radiation.
Midjourney Medical lowers a person into warm water while a ring of ultrasound sensors travels along the body. Real volunteers have been scanned, and academic teams have separately published human whole-body ultrasound tomography.
Signal versus noise
Predictive value depends on how common cancer is in the tested population, not just laboratory accuracy.
This is a teaching example, not the performance of a named test. It assumes 75% sensitivity and 99.5% specificity.
Among people who have the cancer, the proportion whose test is positive.
Among people without the cancer, the proportion whose test is negative.
Among people with a positive result, the proportion who actually have the cancer.
Finding a cancer that would never have caused illness or death during that person’s lifetime.
After a positive result
A screening signal does not equal a cancer diagnosis.
Confirm identity, quality, symptoms, family history and plausible benign causes.
Targeted examination, ultrasound, CT, MRI, endoscopy or another appropriate test.
Tissue pathology remains central for most cancers, with carefully selected exceptions.
Histology, molecular markers and disease extent guide treatment rather than the screening test alone.
Hard-to-detect cancers
There is no validated general-population blood-screening pathway for pancreatic, ovarian or brain cancer. Symptoms and individual risk still determine when imaging and specialist assessment are warranted.
Early pancreatic cancer may release too little signal and cause few specific symptoms. CA 19-9 can support diagnosis or monitoring in some people, but it is not accurate enough for population screening.
Circulating DNA methylation and fragment patterns, proteins, metabolites, extracellular vesicles and combinations with imaging. A useful test must find resectable disease without creating many invasive false-positive investigations.
PANXEON also identified high-grade dysplasia in 64.3% of studied people with high-risk pancreatic cysts. This is an early research signal, not proof that using the test prevents cancer or improves survival. A positive result would need specialist assessment and imaging; a negative result should not replace indicated surveillance.
PANXEON primary studyNCI high-risk surveillance studyCA-125 rises in some ovarian cancers but also in benign conditions, and it can remain normal in cancer. Large trials of CA-125 with transvaginal ultrasound did not reduce ovarian-cancer deaths in average-risk populations.
Longitudinal protein patterns, tumour DNA and methylation, Pap-sample material, extracellular vesicles and multi-analyte panels. New tests must prove a mortality benefit and avoid unnecessary operations after false alarms.
NCI ovarian screening evidenceThe blood-brain barrier and low release of tumour material make routine blood detection particularly difficult. There is no blood test that can rule out a brain tumour. MRI provides the most detailed assessment; CT is faster and often used urgently.
Tumour DNA in blood or cerebrospinal fluid, methylation classifiers, extracellular vesicles, advanced MRI, radiomics and “sonobiopsy” that temporarily increases biomarker release. These remain specialist or research approaches.
NCI brain-tumour liquid-biopsy researchA negative screening or multi-cancer blood test does not overrule symptoms. Tests designed for asymptomatic screening and tests used to investigate a symptomatic person answer different questions.
Research and trials
Look for prospective validation, prespecified thresholds, representative participants, complete follow-up and patient-important outcomes.
Connected evidence guides
Primary evidence