The AI System That Outperformed Expert Radiologists

Google health has recently developed an AI system that can identify breast cancer more accurately than expert radiologists, an early sign assuring the involvement of AI in improving early detection of disease through screening. 

Not only did the AI reduce false positives, in which patients were told they had cancer when they didn’t, but also false negatives in which the cancer is present but not diagnosed. The normal screening process is known to be imperfect, failing to diagnose breast cancer for one in five women. On the other hand, the mammograms diagnose around half of all women of breast cancer every ten years, many of which are false positives, not only causing anxiety, worry, and financial burden on families but also leading to unnecessary treatments that have cost the US more than 4 billion a year. 

Dominic King, the UK head for Google Health recently said that the breakthrough in AI for mammogram screening was “really exciting” as it could be used to detect cancer in earlier stages when the cancer is even harder to identify and diagnose. Throughout the training and testing of the algorithm, a total of 120,000 mammograms were used from the UK and US. This helped the AI outperform specialists by detecting cancers that the radiologists missed in the images while ignoring features they falsely flagged as possible tumors. It then analyses mammograms in three different ways before combing the results to produce an overall risk score. 

The main reason for embarking on this journey and creating the AI was because many radiologists did not think that the screening services available were sustainable relative to the amount of people who needed screening. In 2018, the royal college of radiologists stated that in order to screen all patients effectively and accurately, the UK would need more than 1000 full-time diagnostic radiologists. The aim is not for the AI to replace these radiologists in their jobs, but to get regulatory approval to be a tool that aids radiologists decisions by giving a second opinion, recommending to spend more time looking at certain scans or bringing up cases where the cancer is missed to avoid repeated mistakes. Studies have also shown that the AI model has reduced false positives by 5.7 percent in the US and 1.2 percent in the UK and false negatives by  9.4 percent in the US, and 2.7 percent in the UK.

The next major step is to assess the AI model in real-world conditions, as many experts predict that the current success of the healthcare tool could slip when subjected to mammograms from different screening systems. They will then be evaluated in practice and used in real-time screening with patients, overseen by UK public health agencies that have responsibility for breast screening programs. 

Many tech companies are gaining interest in using their artificial intelligence expertise to aid healthcare, especially creating computer vision algorithms to spot patterns in images and identify any visual signs that are crucial for diagnosis in fields such as pathology, ophthalmology, and dermatology. Google has also already created the Lymph node assistant, a healthcare tool that is 99% accurate at detecting late-stage breast cancer cells and DeepMind is preparing to commercially launch a device that can diagnose complex eye diseases as accurately as specialists. All of these developments will help in aiding doctors to diagnose conditions with increased accuracy in the future and partially ease the burden off the NHS. 

HealthcarePreesha Gehlot