Article

# Help us detect and pre-detect oral cancers

Oral and neck cancer, and why the fight starts in the dental chair

Author: Dr Fady Abboud

## The lesson

Second year of dental school. Radiology class.

The instructor put a panoramic X-ray of one of her patients on the screen and asked: **what do you see?**

The answers came fast. A class II on that molar. An impacted third. Someone checked the sinuses. Good answers. Tooth answers.

I am not usually the one who gets it. That day I said: _the hyoid bone._

She said: continue.

I said there is a radiopacity just above it. Like a small ball. People turned around, _what is he talking about?_

She said: elaborate.

I said it could be a cancer, or something on its way to becoming one.

She said yes. Twice.

Then she said the thing I still repeat:

> You are dentists, but you are medical professionals first. Before you look at the teeth, you verify everything around them: hard tissue, soft tissue, the borders of the film. Patients trust us with their lives, not only their teeth.

That is the whole article. Everything below is the evidence for it.

## The problem is not the cancer. It is the calendar.

389,846 new cases of mouth and oral cavity cancer worldwide in 2022.

This disease is survivable if you find it early. SEER is blunt about it:

Stage at diagnosis

Share of cases

5-year relative survival

Localized

26%

88.4%

Regional (nodes)

54%

69.4%

Distant

13%

36.9%

Three quarters of patients arrive late. Up to half are diagnosed with advanced lesions.

That is not a treatment failure. It is a detection failure, and it happens upstream of the oncologist.

## The person upstream is the dentist

Patients see us more often than they see a physician, and they see us with their mouth open. About 60% of the US population has an annual dental exam. Fewer than 15% of regular dental attendees report ever having had an oral cancer screening.

The exam is not expensive or long: a basic oral cancer examination takes roughly **90 seconds** of looking and palpating.

So why is it skipped? Not laziness. Confidence. In the evidence behind the UK referral guidelines, **85% of general practitioners and 63% of dentists said they did not feel confident examining for oral cancer.**

Sixty-three percent. That is the gap that kills people.

## The pre-signs are boring, which is why they are missed

They do not look like cancer. They look like a nuisance the patient stopped noticing.

-   Persistent unexplained head or neck lump > 3 weeks
-   Oral ulceration or unexplained swelling > 3 weeks
-   Red, or mixed red-and-white patch > 3 weeks
-   Unexplained tooth mobility, not explained by periodontal disease
-   Non-healing extraction socket
-   Persistent unilateral throat discomfort > 4 weeks
-   Ear pain with a normal ear

An unexplained oral ulceration lasting more than three weeks, or a persistent unexplained neck lump, warrants urgent referral within two weeks. Cervical lymphadenopathy, deep cervical or supraclavicular, is itself an indication for urgent referral.

Loose teeth. A socket that will not close. Referred otalgia. These land in a dental chair, not a physician's office.

![Head and neck anatomy showing the hyoid, infrahyoid muscles and common head and neck cancer sites](https://www.sixmonth.ai/assets/articles/head-neck-anatomy.jpg)

The neck a dentist palpates and the neck an oncologist operates on are the same neck.

## The panoramic does not stop at the teeth

This is what my instructor was driving at. The edges of an OPG are not decoration. They are free medical imaging of the neck, and most of us scroll past them.

![Panoramic dental radiograph with the collateral anatomy labelled](https://www.sixmonth.ai/assets/articles/panoramic-anatomy.jpg)

Everything numbered here is outside the dentition. All of it is on every panoramic you take.

In a series of 2,017 panoramics, incidental findings appeared in **26.23%** of images: elongated styloid processes, carotid artery calcifications (8.28%), sinus pathology. Among patients aged 54 and older, carotid calcifications appeared on **21.68%** of panoramics.

And that radiopaque smudge near C3 to C4, above the hyoid, is not always an atheroma. The differential includes **calcified lymph nodes**, sialoliths, phleboliths, tonsilloliths and triticeous cartilage. A calcified node is not a curiosity. It is a question that needs an answer.

Here is the damning number. When a carotid calcification _was_ spotted on a panoramic, only **41.7%** of those patients went on to a medical consultation.

We see it. Then nothing happens.

## AI sees the pixel. It does not remember the patient.

The lesion-detection problem is close to solved in the lab.

-   Pooled across imaging-based studies: sensitivity **0.87**, specificity **0.81**, AUC **0.9758** for oral potentially malignant disorders and oral cancer.
-   On ordinary white-light photographs: **81%** sensitivity for OPMDs, **83.6%** for oral cancer.
-   On panoramics, a CNN detected carotid atheromas at **94.7%** accuracy, **95.7%** sensitivity. Another reached 0.87 recall at 0.97 specificity per patient.
-   For jaw lesions, YOLOv8 separated benign from malignant at **0.90** sensitivity. In an earlier head-to-head, the model matched oral surgeons and general practitioners with no statistically significant difference.

The models have converged. They all see the pixel. Be honest about the limits, accuracy drops sharply from histopathology to clinical photographs, because training sets do not cover the full range of how oral disease actually presents, but the trajectory is not in doubt.

What no classifier does today is **remember the patient**.

A lesion is a pixel event. Cancer is a trajectory: a smoker with a 40-pack-year history, a white patch that was on the lateral tongue eighteen months ago and is slightly redder now, a node that was 4 mm and is now 7 mm, a socket that never closed, a hoarseness typed into a hygienist's free-text field and never followed up.

That is the layer **Sixmonth** is building: detection power fused with the full clinical context of the patient, history, clinical signs, pre-signs, prior radiographs, carried forward as a running probability instead of a one-shot verdict. If any layer sees something shift, at any visit, it gets flagged.

Not a diagnosis. A refusal to let the thing disappear between appointments.

Because the failure mode in oral cancer has never been that nobody looked. It is that somebody looked, once, and nobody looked again.

## The point

We are the profession that sees the mouth on a schedule. That is an enormous, unearned advantage, and we mostly spend it on enamel.

Check the borders of the film. Palpate the neck. Ask about the ulcer that has been there a month.

**Patients trust us with their lives, not just their teeth.**

## References

1.  [Global cancer statistics 2022: GLOBOCAN estimates, Bray et al., CA Cancer J Clin](https://acsjournals.onlinelibrary.wiley.com/doi/10.3322/caac.21834)
2.  [Mouth and oral cancer statistics, World Cancer Research Fund](https://www.wcrf.org/preventing-cancer/cancer-statistics/mouth-and-oral-cancer-statistics/)
3.  [Oral Cavity and Pharynx Cancer, SEER Cancer Stat Facts](https://seer.cancer.gov/statfacts/html/oralcav.html)
4.  [Patients' perceptions of oral cancer screening in dental practice, BMC Oral Health](https://pmc.ncbi.nlm.nih.gov/articles/PMC3540027/)
5.  [Improving Oral Cancer Survival: The Role of Dental Providers](https://pmc.ncbi.nlm.nih.gov/articles/PMC2866626/)
6.  [Head and Neck Cancer, Referral Guidelines for Suspected Cancer (NICE / NCBI Bookshelf)](https://www.ncbi.nlm.nih.gov/books/NBK45770/)
7.  [Rapid referral guidelines, head and neck cancer, Macmillan / NICE](https://www.macmillan.org.uk/healthcare-professionals/cancer-pathways/prevention-and-diagnosis/rapid-referral-guidelines/head-and-neck-cancer)
8.  [Oral cancer red flags for dentists, NHS Wales Dental Referrals](https://www.dental-referrals.nhs.wales/dentists/cancer/)
9.  [Delay in diagnosis of oral cancer: a systematic review](https://pmc.ncbi.nlm.nih.gov/articles/PMC8601633/)
10.  [Diagnostic delay in oral squamous cell carcinoma, Int J Oral Sci](https://www.nature.com/articles/ijos201388)
11.  [Incidental findings detected with panoramic radiography: 2,017 cases](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8214591/)
12.  [Carotid Artery Calcifications on Panoramic Radiographs, IJERPH 2022](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9654866/)
13.  [Prevalence of Calcified Carotid Artery Atheroma on Panoramic Radiographs](https://pmc.ncbi.nlm.nih.gov/articles/PMC5108282/)
14.  [Carotid Artery Calcification on Panoramic Radiography, CVA and CAD association](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11049454/)
15.  [Dental Panoramic Radiographs as Opportunistic Screening for Carotid Calcifications](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12614913/)
16.  [Diagnostic Accuracy of AI in Detecting OPMDs and Oral Cancer: Meta-Analysis 2015 to 2024](https://www.ovid.com/jnls/jpcd/fulltext/10.4103/jispcd.jispcd_116_25~diagnostic-accuracy-of-artificial-intelligence-in-detecting)
17.  [Detection of OPMDs and oral cancer with white light through deep learning, Clin Oral Investig](https://link.springer.com/article/10.1007/s00784-026-07058-5)
18.  [Diagnostic performance of AI using medical diagnostic imaging, Front Oral Health](https://www.frontiersin.org/journals/oral-health/articles/10.3389/froh.2024.1494867/full)
19.  [Automated Detection of Oral Malignant Lesions Using Deep Learning, Oral Diseases](https://pmc.ncbi.nlm.nih.gov/articles/PMC12022385/)
20.  [Deep Learning-Based Detection of Carotid Artery Atheromas in Panoramic Radiographs](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12837473/)
21.  [Deep CNN for screening carotid calcification in dental panoramic radiographs](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10096511/)
22.  [Effectiveness of AI in Automatic Diagnosis of Lesions on Panoramic Radiographs (YOLOv8)](https://pubmed.ncbi.nlm.nih.gov/42298092/)
23.  [Deep Learning for Automated Detection of Cyst and Tumors of the Jaw (YOLOv2 vs clinicians)](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7356620/)