V O L 2 6 I S S U E 6 D E C E M B E R / J A N U A R Y 2 0 1 7 / 1 8 T H E E U R O P E A N F O R U M F O R T H O S E I N T H E B U S I N E S S O F M A K I N G H E A L T H C A R E W O R K RADIOLOGY Psychoradiology: brain MRI-mining 10-16 helps classify ADHD Optoacoustics: the sound of cells The DNA mismatch repair mechanism IT & TELEMEDICINE E-health developments in Spain Building an organisation’s 20-21 digital DNA The key to defeating cancer is knowledge dissemination Machine learning is promising Machine learning is playing an increasing role in computer-aided diagnosis, and Big Data is begin- ning to penetrate oncological imaging. However, some time may pass before it truly impacts on clinical practice, according to leading UK-based German researcher Professor Julia Schnabel, who spoke during the last ESMRMB annual meeting, Mélisande Rouger reports Machine learning techniques are start- ing to reach levels of human perfor- mance in challenging visual tasks. Tools such as the convolutional neural network (CNN or ConvNet), a class of deep neural networks that has been applied to analysing visual imagery, have become instrumental in segmentation tasks. Analysing such huge data is still a challenge However, a number of obstacles remain before adequate image analysis arrives, starting with the huge amount of data analysts must work with, according to Professor Schnabel, computational imaging expert at King’s College London. ‘In imaging, the challenges are that we work in 3-D or 4-D, and we have a lot of features to deal with. If we’re lucky, we deal with hundreds or thousands, but not millions of images, so we don’t have a high number of image data to work with. We have this whole sample size problem.’ The professor also identified the high associated cost and imperfec- tion of training data. Training data may be wrongly labelled, depend- ing on the expertise of the observer. Furthermore, machine learning is resource-intensive: only special- ists and consultants can perform special tasks. ‘I personally couldn’t distinguish a glass nodule from a semi-solid nodule. Only specialist consultants and expert radiologists Julia Schnabel PhD joined King’s College London, in the UK in July 2015 as Chair in Computational Imaging at the Division of Imaging Sciences & Biomedical Engineering, taking over the Directorship of the EPSRC Centre for Doctoral Training in Medical Imaging, which is jointly run by King’s College London and Imperial College London. She is also Visiting Professor in Engineering Science, at the University of Oxford. For segmentation in colorectal cancer with DCE MRI, Schnabel and team extracted variability of normalised signal intensity curves from the dataset using principal component analysis. ‘It’s a very sim- ple technique. We just looked at the mean signal intensity of curves embedded within an over-segmen- tation approach, called superpixels Continued on page 2 www.healthcare-in-europe.com CONTENTS NEWS & MANAGEMENT 1-5 SURGERY RADIOLOGY ULTRASOUND 6-9 10-16 17-19 IT & TELEMEDICINE 20-21 LABORATORY 22-23 DIGITAL PATHOLOGY 24 2-D view and 3-D volumetric rendering of contiguous perfusion-supervoxels for tumour parcellation defined on a 4-D dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) scan based on signal enhancement characteristics (Courtesy: Dr Ben Irving, ISMRM 2017) can do that,’ she pointed out. For a disease such as cancer, the image analysis team needs confirmation from pathology, which is often dif- ficult to obtain. For brain imaging, where different protocols exist, one sees different appearance of the same disease on different image protocols for the same patient and between patients. ‘Disease location and size of these pathologies may vary quite signifi- cantly, and the appearance of dis- ease may be very localised: it may be a very sharp “blob”, or it may be very diffused or infiltrated,’ she explained. Deep neural networks The professor shared practical advice on how to work with CNNs appropriately. She stressed the size of the receptive field of a CNN will determine the amount of informa- tion that will be obtained. ‘The size of patches used is important, since a large receptive field increases com- putation and memory requirements, and (max) pooling leads to loss of spatial information. In contrast , if you use very small patches, they are more susceptible to noise.’ As a solution, Schnabel points to using a multi-scale approach, i.e. having smaller patches operating on small filters and larger ones on larg- er filters, and putting them together in the end. Oncological image analysis brings challenges of its own. Machine learning-based segmentation often degrades when deployed in clini- cal scenarios. This is caused by differences between training and test data due to variations in scan- ner hardware and scanner protocols and sequences, Schnabel explained. ‘There is often an imbalance in the training or test data because of a dif- ferent ratio of healthy vs. pathologi- cal cases, individual patient variabil- ity and individual disease variability – also within the same patient. For example, lesions in the liver usually are a secondary cancer, caused by a primary cancer elsewhere, such as in the colorectum.’ Therefore, it is crucial to choose the appropriate network architec- ture. Currently three models in lit- erature are interesting: DeepMedic, FCN (in Deep Learning Toolkit) and U-Net, which owes its name to its ‘U’ shape. ‘These networks use different approaches and for all these, there is the good, the bad and the ugly,’ she pointed out. An ensemble of multiple models and architectures All three networks use CNN based approaches with good perfor- mance, but there are a lot of meta- parameters – more than input cases –, and the architecture and con- figuration influence performance and behaviour. The ugly part is that chosen models and parameters may be suboptimal of other data and applications. ‘Results and con- clusions may therefore be strongly biased,’ she said. One solution could be to use an ensemble of networks; one such example is ‘EMMA’ (ensemble of multiple models & architectures), for which performance is insensi- tive to suboptimal configuration and behaviour is unbiased by architec- ture and configuration.