About the Project
Chronic Lower Back Pain (CLBP) is one of the major types of pain that is affecting many people around the world. It is estimated that 28.1% of US adults suffer from this illness and 2.5 million of UK population experience this type of pain every day. Most CLBP cases do not happen overnight and they usually develop from a less serious variant of acute lower back pain. An acute type of lower back pain can develop into a chronic one if the underlying cause is serious and left untreated. The longer a person is disabled by back pain, the less chance he or she returns to work and the more health care cost he or she will require. It is therefore important to identify the cause of back pains as early as possible in order to improve the chance of patient rehabilitation. The speediness of early diagnosis can depend on many factors including referral time from general practitioner to the hospital, waiting time for a specialist appointment, time for an MRI scan and time for the analysis result to come out. Currently diagnosing the lower back pain is done by a visual observation and analysis of the lumbar spine MRI images by radiologists and clinicians and this process could take up much of their time and effort. In addition, not all clinicians who see these images could interpret them. This, therefore, rationalizes the need for a new approach to increase the efficiency and effectiveness of the diagnostic imaging reporting.
I’m proposing to develop a new methodology to automatically aid clinicians in performing diagnosis of CLBP. The approach will be based on the current accepted medical practice of manual inspection the MRI scans of the patient’s lumbar spine. There are different possible causes of CLBP and identifying one requires a different procedure than the other. To identify lumbar spinal stenosis for example, the procedure involves identifying boundaries between specific regions in MRI images, measuring distances between specific points in these boundaries and comparing the distances with the range of normal values. Another cause of CLBP, namely intervertebral disc degeneration, can be identified through observing the pixel intensity variations within the disc. Our approach will use deep learning to perform semantic segmentation on these regions and perform boundary delineation process. We will also develop a machine learning model of the segmented intervertebral disc regions to detect the level of degeneration. I have carried out a preliminary work in the form of data gathering and ground truth data development. I used this data to test the feasibility of the methodology by using it to detect lumbar spinal stenosis, one of the possible causes of CLBP. In this project, we will further extend this approach to detect and classify other possible causes of CLBP.
The medical diagnosis of the CLBP is performed through visual observation and analysis of specific slices in both axial and sagittal views of the lumbar spine MRI. To detect lumbar spinal stenosis and intervertebral disc herniation, clinicians locate the boundaries between the Intervertebral Disc (IVD), Posterior Element (PE), Thecal Sac (TS), and Area between Anterior and Posterior (AAP) elements in these slices. They will then measure the shortest distances between these regions, namely the central and foraminal distances, as illustrated in Figure 1. To assess the extent of any intervertebral disc degeneration, clinicians observe the pixel intensity variations within the disc and assign a Pfirrmann grading system to it. To detect any damage to bones, ligaments or joints, clinicians observe the relevant parts of the MRI scan and look for any abnormal changes between adjacent slices.
Our proposed methodology (Al-Kafri et al., 2016) will capture and model these processes as algorithms. It starts with identifying slices in a lumbar spine MRI that are useful and necessary for the detection process. These slices are 2D images in either sagittal or axial view and at certain locations and orientations. The images will be then divided spatially into separate regions, each related to a specific organ by performing image segmentation. The remainder of the process is dependent on the type of cause to be identified.
We will be working with Professor Mohammed Al-Jumaily (Consultant Neuro-Spinal Surgeon, Queen Anne Medical Centre. Harley Street Medical Area. London. And Consultant Neuro-Spinal Surgeon, Dubai Healthcare City. UAE.) for experts’ opinion, data gathering and algorithm implementation. I implemented some parts of the proposed methodology namely the semantic image segmentation and the boundary delineation. I developed ground truth data containing annotated labelled images on the lumbar MRI dataset. This is done by locating four regions of interest, namely the Intervertebral Disc (IVD), Posterior Element (PE), Thecal Sac (TS), and Area between Anterior and Posterior (AAP) elements as illustrated in Figure 2. I have also developed a method of developing and evaluating the suitability of ground truth data for lumbar spine MRI image segmentation on that dataset (Natalia et al., 2018). I did this by carrying out a manual segmentation on these slices multiple times, using a number of participants, and deriving a statistical analysis of the result based on the established machine learning performance metrics.
The most recent progress in this area is the analysis of segmentation and boundary delineation algorithms using SegNet (Badrinarayanan, Kendall and Cipolla, 2017), one of the best semantic segmentation techniques in the literature to date. Our results show accurate delineation of important boundaries of regions in lumbar spine MRI when a SegNet with VGG16 layers (Simonyan and Zisserman, 2015) pre-trained using a subset (more than one million) of non-medical images from the ImageNet database (Deng et al., 2009) was used. The best and worst results of the delineation of important boundaries are shown in Figure 3. The paper describing this work is currently being considered for publication (Al-Kafri et al., no date).
Al-Kafri, A. S., Sudirman, S., Hussain, A., Al-Jumeily, D., Fergus, P., Natalia, F., Meidia, H., Afriliana, N., Sophian, A., Al-Jumaily, M., Bashtawi, M. and Al-Rashdan, W. (2018) ‘Segmentation of Lumbar Spine MRI Images for Stenosis Detection using Patch-based Pixel Classification Neural Network’, in IEEE Congress on Evolutionary Computation. Rio de Janeiro, p. in press.
Al-Kafri, A. S., Sudirman, S., Hussain, A., Al-Jumeily, D., Natalia, F., Meidia, H., Afriliana, N., Al-Rashdan, W., Bashtawi, M. and Al-Jumaily, M. (no date) ‘Boundary Delineation of MRI Images for Lumbar Spinal Stenosis Detection through Semantic Segmentation using Deep Neural Networks’, IEEE Access, p. under review.
Al-Kafri, A. S., Sudirman, S., Hussain, A., Fergus, P. J., Al-Jumeily, D., Al-Jumaily, M. and Al-Askar, H. (2016) ‘A Framework on a Computer Assisted and Systematic Methodology for Detection of Chronic Lower Back Pain Using Artificial Intelligence and Computer Graphics Technologies’, in Lecture Notes in Computer Science, pp. 843–854.
Badrinarayanan, V., Kendall, A. and Cipolla, R. (2017) ‘SegNet: A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation’, IEEE Transactions on Pattern Analysis and Machine Intelligence, 39(12), pp. 2481–2495. doi: 10.1109/TPAMI.2016.2644615.
Csurka, G., Larlus, D., Perronnin, F. and Meylan, F. (2013) ‘What is a good evaluation measure for semantic segmentation?.’, in Proceeding of 24th British Machine Vision Conference, p. 2013.
Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K. and Fei-Fei, L. (2009) ‘Imagenet: A large-scale hierarchical image database’, in Computer Vision and Pattern Recognition, 2009. CVPR 2009. IEEE Conference on, pp. 248–255.
Digital Imaging Group (2018) SpineWeb.
Institute of Medicine (2011) Relieving Pain in America: A Blueprint for Transforming Prevention, Care, Education, and Research. Available at: http://www.paincommunitycentre.org/article/low-back-pain-problem#ref6 (Accessed: 10 September 2018).
Joskowicz, L. (2018) ‘Future Perspectives on Statistical Shape Models in Computer-Aided Orthopedic Surgery: Beyond Statistical Shape Models and on to Big Data’, in Computer Assisted Orthopaedic Surgery for Hip and Knee. Springer, pp. 199–206.
Kim, D. H. and MacKinnon, T. (2018) ‘Artificial intelligence in fracture detection: transfer learning from deep convolutional neural networks’, Clinical radiology. Elsevier, 73(5), pp. 439–445.
Maniadakis, N. and Gray, A. (2000) ‘The economic burden of back pain in the UK’, Pain, 84(1), pp. 95–103. doi: 10.1016/S0304-3959(99)00187-6.
Natalia, F., Meidia, H., Afriliana, N., Al-Kafri, A. S., Sudirman, S., Simpson, A., Sophian, A., Al-Jumaily, M., Al-Rashdan, W. and Bashtawi, M. (2018) ‘Development of Ground Truth Data for Automatic Lumbar Spine MRI Image Segmentation’, in 3rd International Workshop on Internet of Things for Big Data Healthcare, p. in press.
National Office for Statistics (2014) NHS Imaging and Radiodiagnostic activity. Available at: https://www.england.nhs.uk/statistics/statistical-work-areas/diagnostics-waiting-times-and-activity/imaging-and-radiodiagnostics-annual-data/.
NHS (2017) Quarterly Diagnostic Waiting Times, NHS England. Available at: https://www.england.nhs.uk/statistics/statistical-work-areas/diagnostics-waiting-times-and-activity/diagnostics-census-data/.
Simonyan, K. and Zisserman, A. (2015) ‘Very Deep Convolutional Networks for Large-Scale Image Recognition’, in International Conference on Learning Representations.