Signal Attenuation Model Free Classification of Diffusion MR Signals of the Breast Tissue using Long Short-Term Memory Networks

Abstract

Detection and diagnosis of breast cancer from diffusion signals by diffusion-weighted imaging involves in estimation of quantitative metrics by signal attenuation models fitted to the signals. The process suffers from the implementation difficulty of the fitting algorithms and their sensitivity to noise. This study aims development of neural networks to facilitate the classification of the breast tissues from the signals. 37500 diffusion MR signals are synthetically generated for noise-free and noisy conditions by signal-to-noise ratio (SNR) for malignant, benign, and healthy breast tissues. Forty neural networks employing traditional long short-term memory (LSTM) or bidirectional long short-term memory (BiLSTM) blocks up to twenty are trained and tested for the signals using bootstrapping incorporated accuracy analysis. Specificity, sensitivity, and accuracy metrics are computed for the higher performance networks. For noise-free and noisy signals with SNR ≥ 80, networks may achieve excellent sensitivities, specificities, and accuracies (100% at all), but LSTM networks require fewer number of memory blocks. For noisy signals having SNRs ≤ 40, the networks may deliver high to very high sensitivities (74.8-98.3%), specificities (87.4-99.2%), and accuracies (83.2-98.9%) better for malignant and healthy tissues than benign tissue but BiLSTM ones perform slightly better. LTSM networks eliminate the need for any signal decay model while outputting remarkably good performances in the classification of diffusion signals. BiLSTM networks perform slightly better for very noisy conditions. Prospective studies are needed to justify the potential benefits in a clinical setup.

Keywords

• Breast
• classification
• diffusion signal
• long short-term memory network

References

1. [1] G.S. Chilla, C.H. Tan, C. Xu, C.L. Poh. “Diffusion weighted magnetic resonance imaging and its recent trend-a survey.” Quantitative imaging in medicine and surgery, vol. 5, no. 3, 2015, pp. 407-422.
2. [2] L. Tang, and X.J. Zhou. “Diffusion MRI of cancer: From low to high b‐values.” J. Magn. Reson. Imaging, vol. 49, 2019, pp. 23-40.
3. [3] R. Woodhams, S. Ramadan, P. Stanwell, S. Sakamoto, H. Hata, M. Ozaki, S. Kan, Y. Inoue. “Diffusion-weighted imaging of the breast: Principles and clinical applications.” RadioGraphics, vol. 31, 2011, pp. 1059-1084.
4. [4] P. Baltzer, R.M. Mann, M. Iima, E.E. Sigmund, P. Clauser, F.J. Gilbert, L. Martincich, S.C. Partridge, A. Patterson, K. Pinker, F. Thibault et al. “Diffusion-weighted imaging of the breast-a consensus and mission statement from the EUSOBI International Breast Diffusion-Weighted Imaging working group.” Eur Radiol., vol. 30, no. 3, 2020, pp. 1436-1450.
5. [5] D. Le Bihan, and M. Iima. “Diffusion magnetic resonance imaging: What water tells us about biological tissues.” PLoS Biol., vol. 13, 2015, e1002203.
6. [6] M. Zhao, K. Fu, L. Zhang, W. Guo, Q. Wu, X. Bai, Z. Li, Q. Guo, J. Tian. “Intravoxel incoherent motion magnetic resonance imaging for breast cancer: A comparison with benign lesions and evaluation of heterogeneity in different tumor regions with prognostic factors and molecular classification”. Oncology Letters, vol. 16, 2018, pp. 5100-5112.
7. [7] Y. Kim, K. Ko, D. Kim, C. Min, S.G. Kim, J. Joo, and B. Park. “Intravoxel incoherent motion diffusion-weighted MR imaging of breast cancer: association with histopathological features and subtypes.” Br. J. Radiol., vol. 89, no. 1063, 2016, pp. 20160140.
8. [8] N.R. Doudou, Y. Liu, S. Kampo, K. Zhang, Y. Dai, S. Wang. “Optimization of intravoxel incoherent motion (IVIM): variability of parameters measurements using a reduced distribution of b values for breast tumors analysis.” MAGMA, vol. 33, 2020, pp. 273-281.

9. [9] G.Y. Cho, L. Moy, J.L. Zhang, S. Baete, R. Lattanzi, M. Moccaldi, J.S. Babb, S. Kim, D.K. Sodickson, E.E. Sigmund. “Comparison of fitting methods and b-value sampling strategies for intravoxel incoherent motion in breast cancer.” Magn. Reson. Med., vol. 74, no. 4, 2015, pp. 1077-1085.
10. [10] G. Ertas. “Fitting intravoxel incoherent motion model to diffusion MR signals of the human breast tissue using particle swarm optimization.” An International Journal of Optimization and Control: Theories & Applications (IJOCTA), vol. 9, no.2, 2019, pp. 105-112.
11. [11] G. Van Houdt, C. Mosquera, and G. Nápoles. “A review on the long short-term memory model.” Artif. Intell. Rev., vol. 53, 2020, pp. 5929-5955.
12. [12] D. Le Bihan, E. Breton, D. Lallemand, P. Grenier, E. Cabanis, and M. Laval-Jeantet. “MR imaging of intravoxel incoherent motions: application to diffusion and perfusion in neurologic disorders.” Radiology, vol. 161, no. 2, 1986, pp. 401-417.
13. [13] C. Liu, C. Liang, Z. Liu, S. Zhang, B. Huang. “Intravoxel incoherent motion (IVIM) in evaluation of breast lesions: Comparison with conventional DWI.” European Journal of Radiology, vol. 82, no. 12, 2013, pp. e782-e789,
14. [14] S. Hochreiter and S. Jürgen. “Long short-term memory.” Neural computation, vol. 9, no. 8, 1997, pp. 1735-1780.
15. [15] M. Schuster and P.K. Kuldip. “Bidirectional Recurrent Neural Networks.” IEEE Trans. Signal Processing, vol. 45, no. 11, 1997, pp. 2673-2681.
16. [16] D. P. Kingma, and J. Ba. “Adam: A method for stochastic optimization.” arXiv preprint arXiv:1412.6980, 2014.
17. [17] F.E. Harrell, K.L. Lee, D.B. Mark. “Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors.” Stat Med., vol. 15, no. 4, 1996, pp. 361-387.
18. [18] J. Chu, W. Dong, K. He, H. Duan, Z. Huang. “Using neural attention networks to detect adverse medical events from electronic health records.” Journal of Biomedical Informatics, vol. 87, 2018, pp. 118-130.
19. [19] B. Rim, N.J. Sung, S. Min, M. Hong. “Deep learning in physiological signal data: A survey.” Sensors (Basel), vol. 20, no. 4, 2020, pp. 969.
20. [20] P. Nagabushanam, S. Thomas George, S. Radha. “EEG signal classification using LSTM and improved neural network algorithms.” Soft Computing, vol. 24, 2020, pp. 9981-10003.
21. [21] S. Saadatnejad, M. Oveisi, M. Hashemi. “LSTM-Based ECG classification for continuous monitoring on personal wearable devices. IEEE J. Biomed. Health Inform., vol. 24, no. 2, 2020, pp. 515-523.
22. [22] M.H. Hesamian, W. Jia, X. He, P. Kennedy. “Deep learning techniques for medical image segmentation: Achievements and challenges.” J. Digit. Imaging., vol. 32, no. 4, 2019, pp. 582-596.
23. [23] S. Gultekin, A. Saha, A. Ratnaparkhi, J. Paisley. “MBA: Mini-Batch AUC Optimization.” IEEE Transactions on Neural Networks and Learning Systems, vol. 31, no. 12, 2020, pp. 5561-5574.
24. [24] J.H Kim. “Estimating classification error rate: Repeated cross-validation, repeated hold-out and bootstrap.” Computational Statistics & Data Analysis, vol. 53, no. 11, 2009, pp. 3735-3745.