One of the areas where artificial intelligence made most progress in the last decade is image recognition. This area embraces a broad range of application, from recognising a smiling face when you tale a picture with your smartphone (now also spotting closed lids of a blinking eye) to identifying faces to sort your photos on the computer, from people authentication to self driving car ambient awareness… AI can tell, by looking at your face and subtle movements if you are tired, moody, detect emotion, spot a sudden increase in blood pressure, detect an incipient plaque in the carotid arteries and other heart beat, blood circulation anomalies (watch the video – it is an old one, significant progress has been made in the past years).
Because of this progress in image recognition healthcare professionals have got interested in the application of AI to screening medical mammography, radiography, TAC to spot anomalies. More and more evidence is accumulating proving that AI can indeed be a game changer in some specific area like early detection of cancer. A recent study at INRIA has shown that AI can spot lung cancer at a very early stage, a year on average before a trained human eye can detect it. Before raising your expectation too much, be aware that the AI identified several more suspicious spots that after biopsy resulted normal tissue with no cancer or pre-cancer lesions. The number of false positive is still too high to shift to an AI screening but the results are clearly very promising, especially considering that the earlier a lung cancer is detected the better the prognoses. Two years ago Google obtained similar results by training their deep learning algorithm on CT scans of hundreds of (anonymised) people, being able to detect cancerous signs on CT scans one year before the lesions were spotted by doctors. Such an anticipation could lead to a 40% improvement in the survival rate. Google, through its company Verily, is looking into applying AI for early cancer detection in clinical trials.
The application of AI to mammography is now widespread and it is flanking medical eyes in detecting suspicious tissue with very good results. Interestingly, a recent study on Dutch women attitude towards AI mammography screening indicates that women may prefer to have their mammogram analysed by a medical doctor rather than by an AI software. AI is seen as a complement that a doctor may use, not as a substitute.