TB detection using modified Local Binary Pattern features
- 1. Broadband Evolution - Unlocking “The Internet of Things” TB detection using modified Local Binary Pattern features Joshua Leibstein Andre Nel
- 2. Overview • Introduction • The problem • Radiograph normalisation • Lung segmentation • Classification • Preliminary results • Conclusion • Future work
- 3. Introduction • SA TB prevalence third highest globally • 1% of SA population infected annually • People infected with HIV at greater risk • SANAC goal to reduce TB by 50% by 2016 • Image processing (IP) detection scheme
- 4. The Problem Figure 1: Two radiographs captured from the same patient
- 5. • Influencing factors – User settings – Capturing equipment – Post-processing – Analogue digitisation – Poorly trained/overworked radiographers • Large variances between mine worker CXRs • Normalisation required The Problem
- 6. Normalisation • Difference of Gaussian (DoG) energy normalisation Figure 2: The frequency band separation process
- 7. Normalisation Figure 3: Frequency sub-bands 𝐼1 ⋯ 𝐼 𝑛
- 8. • Find weighting 𝜆𝑖 for each sub-band • 𝜆𝑖 = Dataset energy Sample energy (1) • 𝐼 = 𝑖=1 𝑛 𝜆𝑖 𝐼𝑖 (2) Normalisation + + + + + = 𝐼1 𝐼2 𝐼3 𝐼4 𝐼5 𝐼6
- 9. Normalisation Figure 4: Three radiographs from the PHRU set
- 10. Normalisation Figure 5: Energy normalised PHRU radiographs and reference radiographs from the JSRT set
- 11. Segmentation • Active Shape Models (ASMs) – Supervised – Manual segmentations defined by “landmarks” – Sampled “whiskers” – Principal Component Analysis (PCA)
- 12. Segmentation Figure 6: Lungs delineated by a feature vector and perpendicular whiskers
- 13. Segmentation Figure 7: An extracted greyscale profile and resulting normalised gradient
- 14. Segmentation Figure 8: Fitting the mean shape to a sample radiograph
- 15. Segmentation Figure 9: PHRU inputs and ASM segmentations
- 16. Segmentation Figure 10: JSRT inputs, ASM segmentations and SCR ground truths
- 17. Overlap Measure • Ω = TP TP + FP + FN (3) – TP: True Positive – FP: False Positive – FN: False Negative • Ω = 87,5983,986% Figure 11: Overlayed lung masks
- 18. Subdivision Figure 12: Mean lung shape subdivision and RGB masks
- 19. RBF Interpolation • Interpolation of 𝑛-dimensional scattered data • 𝑠 𝐱 = 𝑖=1 𝑛 𝑤𝑖 𝜙 ∥ 𝐱 − 𝐱 𝑖 ∥ (4) Figure 13: Interpolation of the mean shape onto a sample lung segmentation
- 20. RBF Interpolation Figure 14: The original position and the interpolated position of a nodule onto the mean lung shape
- 21. RBF Interpolation Figure 15: Abnormalities warped onto the mean shape
- 22. Probability Measure Figure 16: Probability heat maps for regions 1 - 42 Max PMin P
- 23. Classification • Local Binary Patterns (LBPs) – Texture classification – Uniform patterns Figure 17: Feature vector of a 𝐿𝐵𝑃8,1 𝑟𝑖𝑢2 sample
- 24. Classification • Circularly sampled • Log-likelihood measure • k-Nearest Neighbour Figure 18: Circularly sampled 𝐿𝐵𝑃8,1 𝑟𝑖𝑢2 feature vectors
- 25. Results
- 26. Results
- 27. Results
- 28. Conclusion • Energy normalisation using DoG • ASM lung segmentation • Probability measure • Abnormality classification using LBPs • Integrated detection system
- 29. Future work • Relax segmentation parameters • Interpolation of region boundaries • Smaller training regions • Addition of a shape descriptor • Normalised vs unnormalised • Additional probability measure
- 30. References 1. World Health Organisation, “Global Tuberculosis Report 2012,” 2012. 2. South African National AIDS Council, “National Strategic Plan on HIV, STIs and TB 2012 - 2016,” 2012. 3. S. Basu, D. Stuckler, G. Gonsalves, and M. Lurie, “The production of consumption: addressing the impact of mineral mining on tuberculosis in Southern Africa,” Globalization and Health, vol. 5, 2009. 4. B. Girdler-Brown, N. White, R. Ehrlich, and G. Churchyard, “The burden of silicosis, pulmonary tuberculosis and COPD among former Basotho goldminers,” American Journal of Industrial Medicine, vol. 51, no. 9, pp. 640–647, 2010. 5. B. van Ginneken, C. M. Schaefer-Prokop, and M. Prokop, “Computer-aided diagnosis: how to move from the laboratory to the clinic,” Radiology, vol. 261, no. 3, pp. 719–732, 2011. 6. B. van Ginneken, B. M. ter Haar Romeny, and M. A. Viergever, “Computer-aided diagnosis in chest radiography: a survey,” IEEE Transactions on Medical Imaging, vol. 20, no. 12, pp. 1228–1241, 2001. 7. L. Hogeweg, C. Mol, P. A. de Jong, R. Dawson, H. Ayles, and B. van Ginneken, “Fusion of local and global detection systems to detect tuberculosis in chest radiographs,” Proceedings of the Medical Image Computing and Computer Assisted Intervention, vol. 13, pp. 650–657, 2010. 8. J. Shiraishi, S. Katsuragawa, J. Ikezoe, T. Matsumoto, T. Kobayashi, K. Komatsu, M. Matsui, H. Fujita et al., “Development of a digital image database for chest radiographs with and without a lung nodule: Receiver operating characteristic analysis of radiologists detection of pulmonary nodules,” AJR, vol. 174, pp. 71–74, 2000. 9. R. Philipsen, P. Maduskar, L. Hogeweg, and B. van Ginneken, “Normalization of chest radiographs,” Proceedings of SPIE, vol. 8670, pp. 86 700G–86 700G–6, 2013. 10. T. F. Cootes, and C. J. Taylor, “Active shape models: ‘smart snakes’,” Proceedings of the British Machine Vision Conference, pp. 266–275, 1992. 11. B. van Ginneken, S. Katsuragawa, B. M. ter Haar Romeny, M. Viergever et al., “Automatic detection of abnormalities in chest radiographs using local texture analysis,” IEEE Transactions on Medical Imaging, vol. 21, no. 2, pp. 139–149, 2002. 12. B. van Ginneken, A. F. Frangi, R. F. Frangi, J. J. Staal, B. M. ter Haar Romeny, and M. A. Viergever, “Active shape model segmentation with optimal features,” IEEE Transactions on Medical Imaging, vol. 21, pp. 924–933, 2002. 13. B. van Ginneken, M. Stegmann, and M. Loog, “Segmentation of anatomical structures in chest radiographs using supervised methods: a comparative study on a public database,” Medical Image Analysis, vol. 10, pp. 19–40, 2006. 14. T. Ojala, M. Pietikainen, and T. Maenpaa, “Multiresolution gray-scale and rotation invariant texture classification with local binary patterns,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 24, no. 7, pp. 971–987, 2002. 15. J. Leibstein, A. Findt, and A. Nel, “Efficient texture classification using local binary patterns on a graphics processing unit,” in Proceedings of the twenty-first annual symposium of the pattern recognition association of South Africa, pp. 147–152, 2010.