AI diagnoses prostate cancer with 98% accuracy

Blood sample tube with lab requisition form for PSA test, prostate cancer diagnosis
There is currently no set diagnosis test for prostate cancer. (Getty Images)

Artificial intelligence (AI) could diagnose prostate cancer with up to 98% accuracy, research suggests.

There is currently no set test for diagnosing the disease, with patients typically being identified via a urine sample, rectal examination or blood assessment to check prostate specific antigen (PSA) levels.

Read more: Robot taught to sew up patients after surgery

PSA tests are controversial, with many delivering false positive or negative results.

To determine whether AI could be more successful, scientists from the University of Pittsburgh input over a million tissue sample images from prostate cancer patients.

When tested on 1,600 biopsies from 100 suspected patients, the model picked up on the cancer cases 98% of the time.

AI a ‘major advantage’ in cancer diagnosis

“Algorithms like this are especially useful in lesions that are atypical,” said study author Dr Rajiv Dhir.

“A non-specialised person may not be able to make the correct assessment.

“That’s a major advantage of this kind of system.”

Read more: Coronavirus leaves 2.4 million Britons in ‘cancer care backlog’

Prostate cancer is the most commonly diagnosed form of the disease in the UK, where one in eight men will statistically be told they have the condition at some point in their life.

In the US, one in nine men are expected to be diagnosed in their lifetime.

Unlike cervical or breast cancer, there is no national screening programme for prostate forms of the disease.

While PSA screenings have been suggested, result inaccuracies mean “it has not been proved the benefits would outweigh the risks”.

November Prostate Cancer Awareness month, Man holding Blue Ribbon for supporting people living and illness. Healthcare, International men, Father and World cancer day concept
One in eight men in the UK will statistically be diagnosed with prostate cancer at some point in their life. (Getty Images)

Doctors influenced by ‘biases or past experience’

To test the potential of AI in diagnosing the disease, the Pittsburgh scientists fed their model tissue sample slides of patient biopsies.

Each image was labeled by cancer specialists to “teach” the programme to distinguish between healthy and abnormal tissue.

It was then tested on a separate set of samples taken from 100 consecutive patients seen at the University of Pittsburgh Medical Center for suspected prostate cancer.

Results, published in The Lancet Digital Health, revealed the model had a sensitivity – ability to correctly identify those with the disease – of 98%.

Its specificity, ability to spot those without the disease, came in at 97%.

The scientists noted this is “significantly higher than previously reported for algorithms working from tissue slides”.

Read more: Cancer sufferer is ‘disease free’ thanks to ‘internal radiotherapy’

The model was also the first of its kind to pick up on tumour grading, sizing and invasion of the surrounding nerves, they added.

“These all are clinically important features required as part of the pathology report,” wrote the scientists.

The model even detected cancer in six samples that had slipped by a doctor.

Dr Dhir stressed, however, this could come about if a medic had “simply seen enough evidence of malignancy elsewhere in that patient’s samples to recommend treatment”.

As healthcare increasingly turns to AI to diagnose and treat diseases, the scientists hope more patients will be spotted sooner.

“Humans are good at recognising anomalies, but they have their own biases or past experience,” said Dr Dhir.

“Machines are detached from the whole story. There’s definitely an element of standardising care.”

The scientists stressed, however, that AI will not be replacing doctors any time soon.

When it comes to other cancers, the model would have to be trained to detect different forms of the disease, with tumour markers not being universal across all tissue types.