Computers learn to pinpoint deadly tumours

Health reporter,London(wp/es):

Patients with an increasingly deadly cancer could receive quicker and more effective treatment after a “potentially game-changing” breakthrough using computers to pinpoint the location of tumours.

Experts at University College London Hospitals and Google’sartificial intelligence firm DeepMind Health have taught the machines to analyse CT scans of head and neck cancers.

Early results suggest the computers can read the scans as accurately as human radiographers and oncologists — but in a fraction of the time.

This could allow UCLH — a specialist cancer centre for north and east London and Essex — to treat more patients.

It might also mean patients could receive “adaptive” radiotherapy, where the treatment is repeatedly retargeted as the tumour changes shape or shrinks.

There are over 12,000 diagnoses of head and neck cancers a year in the UK, and 4,000 deaths. 

It is twice as common in men than women and can be hard to treat as tumours may be close to the eyes or other organs. Mortality rates have risen 14 per cent in the past decade.

Today’s breakthrough is expected to lead to the launch of clinical trials by the end of the year.

It follows a DeepMind success at Moorfields Eye Hospital, where computers were found to be as effective as humans at analysing scans for signs of eye disease.

Kevin Sullivan, UCLH’s head of cancer services, said: “It is very early days but the findings so far are really exciting. Having an automated system which could reduce part of radiotherapy planning from hours to minutes is potentially game-changing.

“It could enable us to treat patients more quickly and effectively, with the ultimate goal of improving the outlook for people with head and neck cancer.”

It normally takes clinicians up to four hours to mark up cancerous and healthy tissue on CT scans before radiotherapy, a process called “segmentation”.

Scans can be taken seven to 14 days before treatment, leaving a chance they could be out of date, resulting in the therapy not being as precise as possible.

Researchers used more than 800 scans from 500 UCLH patients to create an algorithm that “directs” the computer’s hunt for cancer. The system was double-checked against scans from the US.

Mr Sullivan said: “This will not replace clinicians. We will need radiographers and doctors to review results of the al-gorithm, and physicists will continue to develop radiotherapy treatment plans. But the algorithm will save time.”