The technology has been tested on two of Hong Kong’s most prevalent cancers – lung cancer and breast cancer –achieving diagnostic accuracies of 91% and 99%, respectively, in durations of between 30 seconds and 10 minutes.
Locating pulmonary nodules in 30 secondsCurrently, in order to diagnose lung cancer at an early stage, most doctors depend on chest CT scans to reveal the small pulmonary nodules, which appear on medical images as shades of small lumps. However, each scan often results in hundreds of images. Assuming the time required to go through each image by naked eye is three seconds – it would take in total around five minutes to finish analysing all images.
Such examinations are time consuming. In addition, since the process relies heavily on doctors’ experience, visual acuity and focus, it might result in misdiagnosis.
Meanwhile, when Prof Pheng-ann Heng and his team at CUHK’s Department of Computer Science and Engineering apply deep learning technology to CT scans, they are able to locate the pulmonary nodules in 30 seconds, with an accuracy of 90%.
“Deep learning makes use of advanced training to improve the sensitivity of the technology. So, that it is able to tackle a major challenge that a naked-eye examination faces – i.e. removing noise and reducing false positives,’ explained Prof Heng. He emphasised that the technology does not only improve clinical diagnosis efficiency, but also reduces errors.
According to Heng, the automated screening and analysis technology has received positive feedback from the medical sector, and Prof Heng is expecting the technology to be widely adopted in the next couple of years. To further improve the technology, he expressed that the team would be working with top hospitals in Beijing – in order to provide solid evidence in support of early diagnosis and treatment of lung cancer.
Detecting metastatic cancer in 5 – 10 minApart from benefiting the clinical diagnosis in lung cancer as the leading cause of death in Hong Kong, the research team has also explored the possibility to speed up the process of diagnosing breast cancer. Since 1990, the number of breast cancer patients in Hong Kong has been consistently on the rise. It is now the most prevalent cancer amongst local women, and the third amongst all cancers.
Currently, to diagnose breast cancer, doctors usually have to extract and examine live tissue samples. After locating the lump through mammograms or MR scans, doctors would extract samples and examine them under the microscope to see if there are signs of tumour and whether the tumour is benign or malignant. A digital histology is of high resolution, often up to one gigabyte in file size (equivalent to a 90-minute high resolution movie.) Similar to the existing practice in diagnosing lung cancer, examining such an image requires a lot of time and energy.
To address this problem, the research team has developed a novel deep cascaded convolutional neural network to process the histopathological images. Making use of a fully convolutional network, the model can efficiently and detect the metastatic cancer with a high-resolution score-map. The entire automated analysis process takes approximately five to 10 minutes, as compared to the 15 to 30 minutes that are required if examined by the naked eye.
The improvement in efficiency has also come at no cost in terms of accuracy. The system has accurately achieved a rate of 98.75%, i.e. 2% higher than analysis conducted by experienced doctors. This indicates that it is an invaluable reference for clinical diagnosis on breast cancer.
The technology may further enhance its accuracy when more data are collected, since the key advantage of artificially intelligent deep learning is that it is able to analyse large quantities of parameters. Prof Heng believes that such system acts as a tireless assistant to the doctors – to quickly identify the source of an illness, enabling a timely and appropriate treatment in the medical sector. MIMS
CUHK explores novel methods to detect early Alzheimer’s Disease
Can AI revolutionise drug pricing and supply the next generation of drug pipelines?
Watson: Transforming healthcare with Artificial Intelligence