CC BY
6REFERENCES
[1] D. Li et al., "Visible-near infrared-II skull optical clearing window for in vivo cortical vasculature imaging and targeted manipulation," Journal of Biophotonics 13(10), e202000142 (2020).
[2] M. He et al., "Aggregation-induced emission nanoprobe assisted ultra-deep through-skull three-photon mouse brain imaging," Nano Today 45((2022).
[3] Z. Hu et al., "In vivo tissue optical clearing assisted through-skull targeted photothrombotic ischemic stroke model in mice," J Biomed Opt 27(6), 065001 (2022).
[4] D. Li et al., "Optical clearing imaging assisted evaluation of urokinase thrombolytic therapy on cerebral vessels with different sizes," Biomedical Optics Express 13(6), 3243-3258 (2022).
[5] Li et al., "A Through-Intact-Skull (TIS) chronic window technique for cortical structure and function observation in mice," eLight 2(1), (2022).
DOI 10.24412/cl-37136-2023-1-134-140
IMAGE-GUIDED NEAR INFRARED SPECTRAL TOMOGRAPHY FOR BREAST CANCER
DIAGNOSIS
JINCHAO FENG1, CHENGPU WEI1, SHUMIN LIN, ZHE LI, KEBIN JIA AND SHUDONG
JIANG2
1Faculty of Information Technology, Beijing University of Technology, China 2Thayer School of Engineering, Dartmouth College, NH, USA
ABSTRACT
Breast cancer is the most common cancer diagnosed among women in the world [1-5], which accounts for 30% of all new cancer diagnosis in women. It is noticeable that breast cancer mortality rates improve dramatically when breast tumor can be detected with imaging tools at an earlier and more treatable stage [6].
Fig. 1 depicts current main imaging modalities for breast cancer screening, diagnosis, and treatment. Although conventional breast imaging modalities may be useful for breast cancer screening or diagnosis [7-9]; however, these approaches are largely dependent on tumor size or structural features of abnormalities and provide comparatively little functional information about molecular-level changes in breast tissue [10], which limits opportunities to detect smaller lesions at early stage before lesions spread regionally and symptoms develop.
Figure 1. Clinical breast imaging modalities.
As a non-invasive and low-cost imaging modality [11], near infrared spectral tomography (NIRST) has a great potential in breast cancer diagnosis and monitoring therapeutic response of neoadjuvant chemotherapy. The aim of NIRST is to use near infrared light to illuminate the breast, and to infer the internal distribution of chromophore concentrations from acquired signals. However, NIRST alone may not replace existing imaging modalities due to its poor spatial resolution (near 4-7 mm), which limits tumor size detection [12].
Since MRI has extremely high sensitivity and provides good structural a prior information about the breast, while NIRST provides functional information about the breast tissue, the integration of NIRST into MRI has attracted much attention in recent years. Fig. 2 shows an MRI-NIRST imaging system designed for breast imaging, and the details about the system can be found in Ref. [13]. To incorporate MRI images into NIRST, the widely used methods are hard [14] or soft [15] priors-based algorithms. However, the main shortcoming of the above two techniques is that both methods require manual segmentation to identify regions. This segmentation may lead to the objectivity in the process of combining images. Additionally, the segmentation step can be time-consuming, and requires sufficient segmentation experience to avoid segmentation bias or
To overcome the shortcoming of segmenting the MRI images, a direct regularization imaging (DRI) method has been developed for NIRST reconstruction [16]. However, T1-weighted (T1W), T2-weighted (T2W), T1-weighted dynamic contrast enhanced (DCE) and diffusion weighted (DW) images are often acquired in standard clinical breast MRI. Among the NIRST absorber concentrations, total hemoglobin (HbT) and water have shown the greatest potential to distinguish malignant from benign breast abnormalities [17]. Considering that total hemoglobin (HbT) values are highly related to DCE contrast, and water values are expected to be highly related to DW contrast, we presented a reconstruction algorithm to encode directly the spatial information derived from DCE and DW MRI into multiple regularization matrices for recovering NIRST HbT and water contents [16]. The motivation for pursuing this algorithm is to improve the accuracy of estimating HbT and water simultaneously. Simulation studies were used to test the approach, and the results are shown in Fig. 3. By combining prior information from dynamic contrast enhanced (DCE) and diffusion weighted (DW) MR images, the absolute bias errors of HbT and water were reduced by 35% and 34% compared to the no-prior case.
Figure 2. The MRI-NIRST imaging system used in the following patient experiments. More details about the
system can be found in Ref. [13].
(a) (b) i. (c) Tl V
(d) (e) (f) (g) (h) HbT 58 45
• • • • • 30 17
• • • • • Water . 76 72 63 54 47
Figure 3. Simulation breast study. (a) MRIT1 image; (b) DCE image; and (c) DW image in the plane z= -9.7 mm from a patient MRI exam. (d) Simulated true HbT (left) and water (right) images; (e-h) reconstructed HbT (p.M) and water (%) images guided by no-prior information (e), DCE prior (f) DW prior (g) and both DCE and DW (DCE-DW) priors (h). Reconstructed images are overlaid on the MRI T1 images. Red arrows in (b) & (c) indicate the tumor. White lines in (d) denote cross sections, through the center of the tumor.
Although DCE MRI is recognized as the most sensitive examination for breast cancer detection, it has a substantial false positive rate and gadolinium (Gd) contrast agents are not universally well tolerated. As a result, alternatives to diagnosing breast cancer based on endogenous contrast are of growing interest. Considering this, endogenous NIRST guided by T2 MRI was evaluated to explore whether the combined imaging modality, which does not require contrast injection or involve ionizing radiation, can achieve acceptable diagnostic performance [18]. Twenty-four subjects were simultaneously imaged with MRI and NIRST system shown in Fig. 2 prior to definitive pathological diagnosis. MRIs were evaluated independently by three breast radiologists blinded to the pathological results. Optical image reconstructions were constrained by grayscale values in the T2 MRI. MRI and NIRST images were used, alone and in combination, to estimate the diagnostic performance of the data. Outcomes were compared to DCE results. Sensitivity, specificity, accuracy, and area under the curve (AUC) of noncontrast MRI when combined with T2-guided NIRST were 94%, 100%, 96%, and 0.95, respectively, whereas these values were 94%, 63%, 88%, and 0.81 for DCE MRI alone, and 88%, 88%, 88%, and 0.94 when DCE-guided NIRST was added, as shown in Fig. 4.
100
80 -
g, 60 H 'I 40
<D IS)
20 -
0 -
100
80
g, 60
(U LTt
-DCE-guided NIRST
-- MRI DCE
MRI DCE + DCE-guided NIRST
40
20
-T2-gu!ded NIRST
-- MRI 12 + DWI
MRI T2 + DWI + T2-guided NIRST
l-1-1-r~
100 80 60 40
20
Specificity {%}
(a)
100 80 60 40 20 Specificity {%}
(b)
Figure 4. ROC curves for a DCE-guided NIRST, DCE MRI, and combined DCE MRI and DCE-guided NIRST, and b T2-guided NIRST, T2 + DWI MRI, and combined T2 + DWI MRI and T2-guided NIRST. DCE dynamic contrast-enhanced, MRI magnetic resonance imaging, NIRST near-infrared spectral tomography.
Note that DRI does not need to segment anatomical images; however, it still needs to model light propagation in tissue, and model errors due to mesh discretization, imperfect boundary conditions, and approximate governing equations, are inevitable in NIRST image reconstruction. Deep learning (DL) has been investigated and shown to improve certain image reconstruction problems [25-26]. Inspired by these developments and with the unique opportunity to incorporate anatomical images into these networks that can further improve NIRST image quality, we developed a DL based algorithm (Z-Net) for MRI guided NIRST image reconstruction [27]. In our approach, segmentation of MRI images and modeling of light propagation are avoided and the concentrations of chromophores of oxy-hemoglobin (HbO), deoxy-hemoglobin (Hb), and water are recovered from acquired NIRST signals guided by MRI images through end-to-end training with simulated datasets. Fig. 5 shows the Z-Net architecture for 2D experiments. Optical signals at 9 wavelengths (661, 735, 785, 808, 826, 852, 903, 912, and 948 nm) and MRI images provide the input to the network. Fig. 6 shows representative recovered images of HbO, Hb and water in the case of a phantom with three inclusions. Compared to reconstructed images by DRI, images recovered with Z-net have values much closer to their ground truths with fewer artifacts. the proposed Z-net method provided accurate recovery of HbO, Hb and water concentrations and errors in recovered values were less than 2% of known values. Compared with DRI, MSE obtained with Z-Net was 92.6%, 99.7% and 91.7% lower for HbO, Hb, and water, respectively. Finally, we applied the Z-net approach to image reconstruction of patient data, and the results are shown in Fig. 7.
In conclusion, MRI image-guided NIRST improves quantification of breast tissue chromophore and breast cancer diagnosis.
MRI image (100*100)
^"Spectral images^ (100*100) HbO
Multispectral signals (16*15*9)
Deconv2x2, BN, RcLU, Conv3x3, BN, RcLU
Max pooling, 2x(Conv3x3, BN, ReLU)
Conv3x3, BN, ReLU, Deconv2x2, BN, RcLU
Concatenation
Figure 5. The Z-Net architecture designed for MRI guided NIRST experiments.
Figure 6. Reconstructed images of HbO, Hb, and water in the case of three inclusions. (a) MRI images, (b) source/detector positions around the phantom, (c)-(e) true and reconstructed images of HbO, Hb, and water,
respectively.
Figure 7. The reconstructed HbO, Hb, and Water images for two patients with the Z-net. The first row shows the Z-net results from a breast cancer patient with a malignant lesion, and the second row is the results from a subject with a benign lesion. Reconstructed images are overlaid on the DCE-MRI.
REFERENCES
[1] B. Chance, S. Nioka, J. Zhang, E. F. Conant, E. Hwang, S. Briest, S. G. Orel, M. D. Schnall, and B. J. Czerniecki, Breast cancer detection based on incremental biochemical and physiological properties of breast cancers: a six-year, two-site study, Acad. Radiol., 12(8), 925-933, 2005.
[2] S. Kukreti, A. E. Cerussi, W. Tanamai, D. Hsiang, B. J. Tromberg, and E. Gratton, Characterization of metabolic differences between benign and malignant tumors: high-spectral-resolution diffuse optical spectroscopy, Radiology, 254(1), 277-84, 2010.
[3] W. Chen, K. Sun, R. Zheng, H. Zeng, S. Zhang, C. Xia, Z. Yang, H. Li, X. Zou, and J. He, Cancer incidence and mortality in China, 2014, Chinese J. Cancer Res., 30(1), 1-12, 2018.
[4] C. E. DeSantis, J. Ma, A. Goding Sauer, L. A. Newman, and A. Jemal, Breast cancer statistics, 2017, racial disparity in mortality by state, CA. Cancer J. Clin., 67(6), 439-448, 2017.
[5] R. L. Siegel, K. D. Miller, and A. Jemal, Cancer statistics, 2019, CA. Cancer J. Clin., 69(1), 7-34, 2019.
[6] B. L. Yun, S. M. Kim, M. Jang, H. S. Ahn, C. Y. Lyou, M. S. Kim, S. A. Kim, T. Song, Y. Yoo, J. H. Chang, and Y. Kim, Does adding diffuse optical tomography to sonography improve differentiation between benign and malignant breast lesions?: observer performance study, J. Ultrasound Med., 34(5), 749-757, 2015.
[7] C. K. Kuhl, Current status of breast MR imaging part2. clinical applications, Radiology, 244(3), 672-691, 2007.
[8] C. K. Kuhl, K. Strobel, H. Bieling, C. Leutner, H. H. Schild, and S. Schrading, "Supplemental breast MR imaging screening of women with average risk of breast cancer, Radiology, 28(2), 361-370, 2017.
[9] S. C. Agner, M. A. Rosen, S. Englander, J. E. Tomaszewski, M. D. Feldman, P. Zhang, C. Mies, M. D. Schnall, and A. Madabhushi, Computerized image analysis for identifying triple-negative breast cancers and differentiating them from other molecular subtypes of breast cancer on dynamic contrast-enhanced MR images: a feasibility study, Radiology, 272(1), 91-99, 2014.
[10] B. W. Pogue, S. Jiang, H. Dehghani, C. Kogel, S. Soho, S. Srinivasan, X. Song, T. D. Tosteson, S. P. Poplack, and K. D. Paulsen, Characterization of hemoglobin, water, and NIR scattering in breast tissue: analysis of intersubject variability and menstrual cycle changes, J. Biomed. Opt., 9(3), 541-52, 2004.
[11] D. A. L. Ighter, J. A. H. Ughes, I. A. I. N. S. Tyles, A. Ndrew, F. Iler, and H. A. D. Ehghani, Multispectral, non-contact diffuse optical tomography of healthy human finger joints, Biomed. Opt. Express, 9(4), 1445-1460, 2018.
[12] P. K. Yalavarthy, B. W. Pogue, H. Dehghani, C. M. Carpenter, S. Jiang, and K. D. Paulsen, Structural information within regularization matrices improves near infrared diffuse optical tomography, Opt. Express, 15(13), 8043-8058, 2007.
[13] M. A. Mastanduno, J. Xu, F. El-Ghussein, S. Jiang, H. Yin, Y. Zhao, K. Wang, F. Ren, J. Gui, B. W. Pogue, and K. D. Paulsen, MR-Guided near-Infrared spectral tomography increases diagnostic performance of breast MRI, Clin. Cancer Res. 21(17), 3906-3912, 2015.
[14] V. Ntziachristos, A. G. Yodh, M. D. Schnall, and B. Chance, MRI-guided diffuse optical spectroscopy of malignant and benign breast lesions, Neoplasia 4(4), 347-354, 2002.
[15] P. K. Yalavarthy, B. W. Pogue, H. Dehghani, C. M. Carpenter, S. Jiang, and K. D. Paulsen, Structural information within regularization matrices improves near infrared diffuse optical tomography, Opt. Express 15(13), 8043-8058, 2007.
[16] J. Feng, S. Jiang, J. Xu, Y. Zhao, B. W. Pogue, and K. D. Paulsen, Multiobjective guided priors improve the accuracy of near-infrared spectral tomography for breast imaging, J. Biomed. Opt. 21(9), 090506, 2016.
[17] H. Soliman, A. Gunasekara, M. Rycroft, J. Zubovits, R. Dent, J. Spayne, M. J. Yaffe, and G. J. Czarnota, Functional imaging using diffuse optical spectroscopy of neoadjuvant chemotherapy response in women with locally advanced breast cancer, Clin. Cancer Res. 16, 2605-2614, 2010.
[18] J. Feng, J. Xu, S. Jiang, H. Yin, Y. Zhao, J. Gui, K. Wang, X. Lv, F. Ren, B. W. Pogue and K. D. Paulsen, Addition of T2-guided optical tomography improves noncontrast breast magnetic resonance imaging diagnosis, Breast Cancer Research 19:117, 2017.
[19] G. Yang, S. Yu, H. Dong, G. Slabaugh, P. L. Dragotti, X. Ye, F. Liu, S. Arridge, J. Keegan, Y. Guo, and D. Firmin, DAGAN: deep dealiasing generative adversarial networks for fast compressed sensing MRI reconstruction, IEEE Trans. Med. Imaging 37(6), 1310-1321, 2018.
[20] K. H. Jin, M. T. McCann, E. Froustey, and M. Unser, Deep convolutional neural network for inverse problems in imaging, IEEE Trans. Image Process. 26(9), 4509-4522, 2017.
[21] H. Lan, D. Jiang, C. Yang, F Gao, and F. Gao, Y-Net: hybrid deep learning image reconstruction for photoacoustic tomography in vivo, Photoacoustics, 20, 100197, 2020.
[22] Y. Gao, K. Wang, Y. An, S. Jiang, H. Meng, and J. Tian, Nonmodel-based bioluminescence tomography using a machine-learning reconstruction strategy, Optica 5(11), 1451-1454, 2018.
[23] S. Sabir, S. Cho, Y. Kim, R. Pua, D. Heo, K. H. Kim, Y. Choi, and S. Cho, Convolutional neural network-based approach to estimate bulk optical properties in diffuse optical tomography, Appl. Opt. 59(5), 1461-1670, 2020.
[24] H. Wang, N. Wu, Z. Zhao, G. Han, Jun Zhang, and J. Wang, Continuous monitoring method of cerebral subdural hematoma based on MRI guided DOT, Biomed. Opt. Express 11(6), 2964-2975, 2020.
[25] J. Yoo, S. Sabir, D. Heo, K. H. Kim, A. Wahab, Y. Choi, S. Lee, E. Y. Chae, H. H. Kim, Y. M. Bae, Y. Choi, S. Cho, and J. C. Ye, Deep learning diffuse optical tomography, IEEE Trans. Med. Imaging 39(4), 877887, 2020.
[26] Y. Takamizu, M. Umemura, H. Yajima, M. Abe, and Y. Hoshi, Deep learning of diffuse optical tomography based on time-domain radiative transfer equation, arXiv:2011.12520, 2020.
[27] J. Feng, W. Zhang, Z. Li, K. Jia, S. Jiang, H. Dehghani, B. W. Pogue, and K. D. Paulsen, Deep-learning based image reconstruction for MRI-guided near-infrared spectral tomography, Optica, 9(3), 264-268, 2022.
