We investigate in vivo detection of mammary tumors in a rat model using autofluorescence imaging in the red and far-red spectral regions. The objective was to explore this method for non-invasive detection of malignant tumors and correlation between autofluorescence properties of tumors and their pathologic status. Eighteen tumor-bearing rats, bearing eight benign and seventeen malignant tumors were imaged. Autofluorescence images were acquired using spectral windows centered at 700-nm, 750-nm and 800-nm under laser excitation at 632.8-nm and 670-nm. Intensity in the autofluorescence images of malignant tumors under 670-nm excitation was higher than that of the adjacent normal tissue. whereas intensity of benign tumors was lower compared to normal tissue.
© 2006 Optical Society of America
Breast cancer is the most frequent and fatal cancer diagnosed in women. The lifetime risk for development of breast cancer in American women is estimated to be 1 in 8 . Although the survival rate has improved in recent years, the number of deaths linked to breast cancer remains high; for the year 2004, it is estimated that over 40,000 women have died of breast cancer in the United States . Mammography and physical examination are the methods currently used to screen women for breast cancer. Although these methods have a high sensitivity (>75 %) , only one of four detected lesions will prove to be cancer at pathology. The remaining tumor masses are benign or inflammatory. It is therefore appropriate to search for a fast, non-invasive, and inexpensive method, such as optical imaging, that could be the next step to characterize tumors based on their physical and chemical properties and thus lead to the improved differentiation of benign from malignant lesions.
The use of light for cancer diagnosis dates back to the 1920s, when transillumination was used to investigate breast cancer . This technique was named diaphanography, but its low sensitivity and specificity limited its clinical usefulness. With progress in photonic technologies, mathematic modeling of light propagation, and increased knowledge of the photophysical properties of tissues, optical imaging (also referred to as optical mammography) has evolved to become a promising imaging tool for oncological applications of the breast [5–7].
The utilization of native optical “signatures” associated with the way tissue components interact with light has also been proven to be a very promising approach for tissue characterization in real time. Changes in the spectroscopic properties of pathological specimens including diverse cancers originating in many organs such as skin, gastrointestinal tract and oral mucosa have stimulated a great deal of interest for its potential application for the detection and treatment of cancer . In the case of breast cancer, Raman spectroscopy [9–11] and light scattering spectroscopy [12,13] have demonstrated potential for the differentiation of breast tissue components. More recent studies using fresh human tissue specimens containing normal and cancer tissue from various body organs (including breast) have indicated the potential of NIR autofluorescence imaging under long wavelength excitation for cancer detection [14–16]. Zhang et al demonstrated that this signal has a lifetime on the order of 1 ns, thus it is due to emission by a tissue chromophore  but, to the best of our knowledge, an exact identification has not yet been made. It has been hypothetized that porphyrins may be the fluorophore giving rise to changes in the NIR autofluorescence intensity in cancer tissue [14,15,17]. This change may be the result of a change in the production of porphyrins in neoplasia due to change in the heme-biosynthetic pathway.
In this study, we investigate NIR autofluorescence imaging for noninvasive detection of subsurface tumors in vivo using a rat model of chemically-induced mammary tumors. This living rodent model was selected because it produces a range of tumors that span the whole pathologic spectrum, such as those observed in human breast masses. A second objective was to evaluate possible correlation of the NIR autofluorescence intensity arising from tumors with their pathologic status.
2. Experimental arrangement
2.1 Tumor model
The study was conducted with the approval of the Committee for Animal Research of the University of California San Francisco, and conformed to the guidelines of the National Institutes of Health for the care and use of laboratory animals.
Tumors were chemically induced in eighteen female Sprague Dawley rats by injecting a single dose of N-ethyl-N-nitrosurea (ENU) intra-peritoneally. Tumors developed in the mammary gland of the rat over a period of 5 to 15 months. Typically, the tumor pathologies mimic the complete range of tumor pathologies encountered in humans, from benign fibroadenomas to highly malignant carcinomas . High doses tend to induce fast-growing malignant tumors, whereas lower doses tend to induce slower-growing benign tumors. To obtain a range of pathologies and tumor grades, different doses of chemical were used, ranging from 90 to 250 mg/kg. Forty tumors in total were imaged and analyzed by pathology, when tumor diameters reached at least 1-2 cm.
2.2 Optical spectroscopic imaging
A schematic diagram of our experimental setup is shown in Fig. 1. Our custom imaging system incorporated an OPO laser pumped at 532 nm by the second harmonic of an Nd:YAG laser. The temporal width of the laser pulses was ≈5 ns and the spectral linewidth was ≈1 nm. This laser operates at a repetition rate of 20 Hz and provided tunable operation from 660 to 970 nm, permitting coverage of the far red and near infrared spectral region. It also included a Helium-Neon laser emitting at a 632.8-nm wavelength. The laser light was transferred into the enclosed imaging compartment of the system using a fiber bundle. The central portion of the diverging laser beam was used to photo-excite the region of interest in the animal in order to acquire the autofluorescence image. A liquid nitrogen-cooled CCD camera was used to capture the images. A laptop computer remotely operated the system.
Prior to imaging, rats were anesthetized with a 20 mg/kg intra-peritoneal dose of pentobarbital, and shaved of their hair in the zone of the tumor and adjacent normal mammary tissue. Each animal was then placed on a trough so as to have the tumor facing the imaging system. An initial image was acquired under ambient light using the same spectral window used for the autofluorescence images. This image was used as an anatomical reference of the tumor position.
For the acquisition of the autofluorescence images, a set of excitation laser wavelengths and emission spectral bands was chosen according to previous reports of tumor autofluorescence [14–16]. More specifically, the excitation wavelengths were either 632.8 nm when using the He-Ne laser, or 670 nm when using the OPO laser. The monochromatic nature of the excitation was ensured by placing a narrow-band filter centered at the laser wavelength (10-nm in bandwidth) at the output of the fiber used to transport the laser excitation as shown in Fig. 1. The spectral bands used to acquire the autofluorescence images were selected using two optical filters, a narrow band filter (with a bandwidth of 40-nm at full width at half maximum) in combination with a long-wavelength-pass filter (to assist achieve complete extinction of the excitation laser light). These filters were positioned in front of the CCD camera. The imaging spectral bands (determined by the narrow band filters) were centered at 700 or 750 nm when using the 632.8 nm laser, and 750 or 800 nm when using the 670 nm wavelength. The narrow band filters were combined with a 670 nm long pass filter when using the 632.8 nm excitation, and a 715 nm long pass filter when using the 670 nm laser wavelength. The detection spectral bands were chosen at least 50 nm higher than the excitation, so as to efficiently exclude the excitation laser light during detection. We confirmed that there was no leakage of the excitation light through the imaging filters by acquiring an image of a non-fluorescing metallic target object for each combination of filters used at the output of the excitation fiber and in front of the imaging lens. This metallic target had unpolished (rough) surfaces so that a strong specular reflection component from the laser beam was collected by the imaging optics. In all cases, this object was not visible for integration times used in this experiment, ensuring that the images recorded from the experimental animals were due to autofluorescence only of the experimental animal. The image acquisition time was one second for all images under average laser power of 30 mW, corresponding to approximately 0.5 mW/cm2 within the imaged area.
2.3 Image analysis
The exact spatial map of the illumination intensity was acquired by recording the fluorescence image of a high quality sheet of paper. Regions of interest (ROI) were placed on the tumor and the adjacent normal mammary tissue, to measure mean signal intensity, and yield a quantitative measure of tumor-to-normal tissue contrast, using the WinView/32 software (Roper Scientific, Inc., Tucson, AZ). The WinView/32 software allowed drawing of rectangular regions of interest placed in the center of the tumor. The image was composed of 512×512 pixels and its size was 13.5 cm2, yielding a spatial resolution of 0.26 mm.
ROI were placed on tumor and normal tissue on the autofluorescence images, guided by the difference of signal intensities between tumor and normal tissue. In rats with several tumors, the autofluorescence intensity was measured on each tumor. To make sure that the possible inhomogeneity of illumination or the angle of the tissue with the laser illumination or the CCD camera did not influence the measures, a ROI was placed on adjacent normal tissue for each tumor. The tumor-to-normal tissue ratio (T/N) was calculated as the ratio of the average signal intensities of the tumor over the adjacent normal tissue. Analysis of the images was performed prior to knowledge of the pathological results.
2.4 Pathological analysis
At the end of each imaging protocol, animals were sacrificed by a lethal intravenous dose of pentobarbital followed by bilateral thoracotomies. Tumors were resected, fixed in 10 % buffered formalin, processed into paraffin and sectioned. Conventional hematoxylin and eosin staining was performed for microscopic assessments of tumor morphology and tumor grading according to the Scarff-Bloom-Richardson (SBR) method. This method grades malignant tumors according to three characteristics: glandular formation, nuclear pleomorphism, and mitotic activity (scores 3-9 reflecting progressive grades of malignancy) .
2.5 Statistical analysis
Tumor-to-normal tissue ratio values (T/N) were compared between the benign and malignant groups of tumors by performing non-parametric Mann-Whitney tests using the statistical software Statview (SAS Institute Inc., Cary, NC). A P value less than 0.05 was considered statistically significant. The best combination of excitation/detection filters was chosen as that which yielded the most statistically significantly different results between the T/N ratios of tumor and normal soft tissues. Sensitivity, specificity, positive and negative predictive values were calculated using a criteria of T/N < or > 1 to differentiate between benign and malignant tumors. An ANOVA analysis was performed to compare the estimated OI-derived autofluorescence parameter T/N, with the histological tumor SBR score, the ENU injection dose, and the incubation period before the tumors appeared.
3. Experimental results
3.2 Tumor characterization
From the forty mammary tumors examined, three were excluded because they were not breast tumors (one was a sarcoma, another was a pilo-sebaceous tumor, and the last an atypical carcinoma). Another twelve (5 benign, 7 malignant) were excluded for two main reasons. A number of animals had patches of fine hair remaining after shaving which we were not able to remove due to the risk of skin irritation. Indeed, during the experiments we noticed that local inflammation also increased the autofluorescence signal and risked yielding false positives. In addition, in five animals a pus-like collection was discovered at dissection near the tumors corresponding to areas of high NIR fluorescent signal in the images. In each of these cases, it was the author’s opinion (after considering the entire set of experimental results and behaviors observed during the experiments) that the signal intensity measured could not be linked “without a doubt” to the signal content of the tumor or normal tissues.
From the twenty five remaining tumors included in this study (in fifteen rats), seventeen were malignant with Scarff-Bloom Richardson scores ranging from 3 to 6. The eight other tumors were benign fibroadenomas. The malignant tumors developed on average 6½ months after ENU administration, while the benign tumors tended to develop more slowly, on average 10 months after exposure. It can be noted that none of the tumors included in this study demonstrated necrosis on pathological examination. All tumors were palpable, as they were situated directly beneath the skin. Tumor sizes were not statistically different between malignant tumors (mean size: 27 ± 9 mm) and benign tumors (mean size: 25 ± 13 mm). The latter however had much more dispersed tumor sizes, with either small or large tumors, but no intermediate size.
3.2 Tumor-to-normal signal intensity ratio (T/N)
The size of the regions of interest varied according to the size of the tumors: mean tumor ROI sizes were 7993 ± 8089 pixels. For normal tissue, region sizes were always approximately the same size, arbitrarily drawn as roughly a 1 cm2 square (≈ 1500 pixels).
Analysis of the experimental results showed that the 670/800 nm combination yielded statistically significant differences between benign and malignant tumor-to normal signal intensity ratios, as well as the best qualitative (visual) contrast between the tumors and the normal tissue. No statistically significant differences were observed under 632.8-nm excitation. The results are summarized in Table 1. The individual results for each tumor obtained under 670-nm excitation and 800-nm detection are presented in Table 2.
Thirteen of the seventeen cancers appeared visually brighter than adjacent normal mammary soft tissue, with higher cancer/normal tissue signal intensity ratios (T/N>1) for the 670/800-nm combination. Though ratios of malignant tumors were only a few percentage points for certain tumors, it is noteworthy that even in these cases, the difference in fluorescence was visible to the eye (for example the second tumor in Fig. 2 with a T/N ratio of 1.01). The four other malignant tumors (numbers 15, 20, 21 and 24) had signals that were inferior to the signal of adjacent normal mammary soft tissues. Fig. 2 demonstrates a typical example showing the image of three malignant tumors within the same rat clearly visible in the autofluorescence image due to their higher intensity compared to than of the adjacent normal tissue. In this image, the section of the animal that was not shaved from its hair appears as the area of highest intensity due to the fact that the rat’s hair is highly emissive in the NIR spectral region. Of the eight benign tumors, six were darker than the adjacent normal tissue (T/N<1), in sharp contrast to the malignant tumors. A typical example is shown in Fig. 3. The other two benign tumors appeared brighter than the normal tissue (numbers 17 and 18).
Figure 4 displays the T/N signal intensity ratios of benign and malignant tumors as a function of their size. The intensity ratios of the malignant tumors have a signal that seems independent of size. Indeed, whether small or large, they appear to be more fluorescent than normal mammary tissue. The signal intensity of the four biggest benign tumors, on the other hand, seem to be negatively correlated to size. No definite conclusion can be made, however, due to the small number of benign tumors.
3.3 Statistical analysis
The experimental results in this study of twenty-five induced mammary tumors in rodents demonstrate that optical imaging using the near-infrared native fluorescence was able to non-invasively detect and image most of the malignant tumors. Furthermore, the statistical analysis suggests that this method may help differentiate benign from malignant lesions. More specifically, the mean values under 670 nm excitation of the signal intensity ratios of tumor-to-normal-mammary-tissue (T/N) were 0.83 for the benign tumors, and 1.14 for the malignant tumors for images acquired using the 800±20 nm filter (detection spectral range). These values differed significantly (P=0.009). The mean values under 670 nm excitation and 750±20 nm detection were 0.82 for the benign tumors, and 1.12 for the malignant tumors with a statistically borderline value of P=0.10. Under 632.8 nm excitation, no difference was observed between the normal and tumor (benign or malignant) tissue in the NIR autofluorescence images.
An ANOVA showed no correlation between the T/N ratio and ENU dose or incubation period (respectively P=0.8, 0.5 and 0.5).The analyses failed to show statistically significant levels between the T/N ratios and the SBR tumor scores, whether the analysis included the benign tumors (SBR scored at 3) or not (respectively P=0.16, and 0.48).
To evaluate the possibility that NIR autofluorescence imaging may differentiate benign from malignant tumor, we considered that tumors with higher autofluorescence than adjacent normal tissue (T/N>1) were malignant, and tumors with lower autofluorescence than adjacent normal tissue (T/N<1) were benign. A statistical analysis with these parameters under 670-nm excitation and 800-nm detection yielded a sensitivity of 76 %, a specificity of 75 %, a positive predictive value of 87 %, and a negative predictive value of 60 % (Table 3).
TN is true negatives, TP true positives, FN false negatives, FP false positives, Se sensitivity, Sp specificity, PPV positive predictive value and NPV negative predictive value.
This work has been motivated by recent reports describing ex vivo the detection of numerous cancer types such as breast, kidney and bladder using the NIR autofluorescence under long wavelength excitation [14–16]. It has been hypothesized that emission by porphyrins present at different concentrations in tumor tissue compared to corresponding normal tissue gives rise to difference in the NIR emission intensity providing for visualization of the tumor in an imaging arrangement. Increased porphyrin production has been observed in multiple cancer types including cancers as varied as breast, colon, endocrine, kidney, liver or lymphomas [20–22]. Although the absorption peaks of porphyrins are in the visible and near ultraviolet spectral range, their absorption spectrum extends up to 670 nm , which is the longest excitation wavelength used in this work. However, in breast cancers a direct correspondence between autofluorescence in the near-infrared and porphyrin content remains to be demonstrated. Other possible explanations of NIR-fluorescence observed in biological tissues are necrosis and bacterial infection . Pathology did not show either of these conditions to be present in our series of tumors. Based on these previous reports, we hypothesized that enhanced emission by the tumors in our experimental animals may provide the optical signature for noninvasive detection and imaging of subsurface tumors. The choice of excitation in the red and far-red spectral region was guided by the need to achieve most efficient excitation of the fluorescing biomolecule (porphyrin?) while avoiding the excitation of other tissue fluorophores. In particular, the absorption coefficient by hemoglobin, an important source of absorption in tissues, is minimized for wavelengths longer than about 650 nm.
The quantification of the autofluorescence intensity of the lesions was performed using the tumor-to-normal tissue signal intensity ratio for two reasons. First, ratios allowed for comparison among different animals/tumors. The autofluorescence intensity of the same tissue component varies between different animals (or humans) which necessitates some kind of normalization procedure through comparison between tissue components of the same subject . Secondly, this T/N ratio closely parallels the qualitative image contrast relied upon routinely by diagnostic imagers for tumor assessments in a clinical context.
The analysis of the experimental results summarized in Table 1 reveal a trend that may help explain the observed behaviors based on tissue optics concepts. Given that the tumors are located below the surface and that the excitation light must penetrate through the skin, the 670 nm excitation will more efficiently reach and penetrate into the tumor due to lower absorption and scattering compared to 632.8 nm excitation. Absorption by blood in the vascular network of the tumor may be the main mechanism providing for a difference in the excitation conditions under 632.8 and 670 nm illumination (the molar extinction coefficient of hemoglobin at 633 nm is twice that at 670 nm). Both malignant and benign tumors exhibit high vascularity. This lower absorption at 670 nm may be essential in providing more efficient excitation of the tumor to increase the relative intensity detected signal by the imaging system (compared to signal arising from the skin and non-tumor components of the animal). Similarly, the longer emission wavelength (at 800±20) will be more efficient in reaching the detection system due to lower scattering by the tissue (compared to the shorter imaging wavelength at 750±20 nm) thus providing a better image contrast and lower P value as indicated from the statistical analysis of our experimental results.
Due to the small number of benign tumors, absolute conclusions cannot be inferred from this sole experiment. However, the observed decrease in fluorescence signal intensity for the large benign fibroadenomas was a salient result in our experiment. This model of benign tumors has been shown in previous reports to be equally or less vascular than the malignant carcinomas [25,26]. Consequently, the lower signal observed in benign tumors may not be attributed to higher absorption of (excitation and emission) light by hemoglobin or deoxy-hemoglobin. However, benign tumors exhibit higher content in fibroblasts and collagen which could alter the optical properties and lead to lower fluorescence. One may also consider the possibility that the NIR fluorescing fluorophore is present at lower concentrations in benign tumors compared to malignant tumors. The limited results from benign tumors included in this work also suggest a dependence of the NIR autofluorescence intensity on the size of the tumors. It is evident that a more detailed investigation of the benign tumor intensity is required before a definite conclusion may be drawn regarding the ability of this technique to differentiate malignant from benign tumors and at what stage of development.
Near-infrared fluorescence imaging revealed statistically significant differences of an inherent property (NIR autofluorescence) among normal tissue, benign fibroadenomas and malignant adenocarcinomas in a rat model of breast tumors. This study suggests that NIR autofluorescence spectroscopy may help devise methods for in vivo characterization of breast tumors as well as improve specificity of lesion assessment compared to existing modalities. Currently, x-ray mammography and ultrasound offer have good sensitivity, but low specificity  (only 20-25 % of lesions undergoing biopsy are ultimately identified as cancer). This work also demonstrated that NIR autofluorescence imaging can detect and image subsurface lesions when differences in the emission intensity between normal and tumor components exist.
This work was sponsored by a grant from the University of California Office of the President CLC-01-28. This work was performed in part at Lawrence Livermore National Laboratory under the auspices of the U.S. Department of Energy under Contract W-7405-Eng-48. This research is supported in part by the Center for Biophotonics, a National Science Foundation Science and Technology Center, managed by the University of California, Davis, under Cooperative Agreement No. PHY 0120999.
References and links
1. National Cancer Institute, “Cancernet resource” (National Cancer Institute, 2003), http://cis.nci.nih.gov/fact/5_6.htm.
2. National Cancer Institute, “Surveillance, Epidemiology, and End Results: Estimated new cancer cases and deaths for 2004” (National Cancer Institute, 2004), http://seer.cancer.gov/cgi-bin/csr/1975_2001/search.pl#results.
3. T. M. Kolb, J. Lichy, and J. H. Newhouse, “Comparison of the performance of screening mammography, physical examination, and breast US and evaluation of factors that influence them: an analysis of 27,825 patient evaluations,” Radiology 225, 165–75 (2002). [CrossRef] [PubMed]
4. M. Cutler, “Transillumination as an aid in the diagnosis of breast lesions.,” Surg. Gynecol. Obstet. 48, 721–729 (1929).
5. X. Gu, Q. Zhang, M. Bartlett, L. Schutz, L. L. Fajardo, and H. Jiang, “Differentiation of cysts from solid tumors in the breast with diffuse optical tomography,” Acad. Radiol. 11, 53–60 (2004). [CrossRef] [PubMed]
6. D. B. Jakubowski, A. E. Cerussi, F. Bevilacqua, N. Shah, D. Hsiang, J. Butler, and B. J. Tromberg, “Monitoring neoadjuvant chemotherapy in breast cancer using quantitative diffuse optical spectroscopy: a case study,” J. Biomed. Opt. 9, 230–8 (2004). [CrossRef] [PubMed]
7. V. Ntziachristos, A. G. Yodh, M. Schnall, and B. Chance, “Concurrent MRI and diffuse optical tomography of breast after indocyanine green enhancement,” Proc. Natl. Acad. Sci. U. S. A. 97, 2767–72. (2000). [CrossRef] [PubMed]
8. G. A. Wagnieres, W. M. Star, and B. C. Wilson, “In vivo fluorescence spectroscopy and imaging for oncological applications,” Photochem. Photobiol. 68, 603–32. (1998). [PubMed]
10. A. S. Haka, K. E. Shafer-Peltier, M. Fitzmaurice, J. Crowe, R. R. Dasari, and M. S. Feld, “Identifying microcalcifications in benign and malignant breast lesions by probing differences in their chemical composition using Raman spectroscopy,” Cancer Res. 62, 5375–80 (2002). [PubMed]
11. R. Manoharan, K. Shafer, L. Perelman, J. Wu, K. Chen, G. Deinum, M. Fitzmaurice, J. Myles, J. Crowe, R. R. Dasari, and M. S. Feld, “Raman spectroscopy and fluorescence photon migration for breast cancer diagnosis and imaging,” Photochem. Photobiol. 67, 15–22 (1998). [CrossRef] [PubMed]
12. S. G. Demos, H. Savage, A. S. Heerdt, S. Schantz, and R. R. Alfano, “Time Resolved Degree of Polarization for Human Breast Tissue,” Optics Commun. 124, 439 (1996). [CrossRef]
13. D. W. Chicken, A. C. Lee, G. M. Briggs, M. R. S. Keshtgar, K. S. Johnson, D. D. O. Pickard, I. J. Bigio, and S. G. Bown, “Optical biopsy: A novel intraoperative diagnostic tool to determine sentinel lymph node status instantly in breast cancer,” Breast Cancer Res. Treat. 82, S172–S174 (2003).
14. S. G. Demos, R. Gandour-Edwards, R. Ramsamooj, and R. DeVere White, “Spectroscopic detection of bladder cancer using near-infrared imaging techniques,” J. Biomed. Opt. 9, 767–71 (2004). [CrossRef] [PubMed]
16. G. Zhang, S. G. Demos, and R. R. Alfano, “Far-red and NIR spectral wing emission from tissues under 532-nm and 632-nm photo-excitation,” Lasers Life Sci. 9, 1–16 (1999).
17. M. Inaguma and K. Hashimoto, “Porphyrin-like fluorescence in oral cancer: In vivo fluorescence spectral characterization of lesions by use of a near-ultraviolet excited autofluorescence diagnosis system and separation of fluorescent extracts by capillary electrophoresis,” Cancer 86, 2201–11. (1999). [CrossRef] [PubMed]
18. G. Stoica and A. Koestner, “Diverse spectrum of tumors in male Sprague-Dawley rats following single high doses of N-ethyl-N-nitrosourea (ENU),” Am. J. Pathol. 116, 319–26 (1984). [PubMed]
19. R. Scarff and H. Torloni, “Histological typing of breast tumors,” in World Health Organization, (Geneva, 1968), pp. 13–20.
21. L. Rasetti, G. F. Rubino, L. Tettinatti, and G. W. Drago, “Porphyrin porphobilinogen and amino ketone levels in tumor tissue,” Panminerva Med. 7, 105–110 (1965).
22. B. Zawirska, “Comparative porphyrin content in tumors with contiguous non-neoplastic tissues,” Neoplasma 26, 223–9. (1979). [PubMed]
23. T. Theodossiou and A. J. MacRobert, “Comparison of the photodynamic effect of exogenous photoprotoporphyrin and protoporphyrin IX on PAM 212 murine keratocytes.,” Photochem. Photobiol. 76, 530–537 (2002). [CrossRef] [PubMed]
25. H. Daldrup, D. M. Shames, M. Wendland, Y. Okuhata, T. M. Link, W. Rosenau, Y. Lu, and R. C. Brasch, “Correlation of dynamic contrast-enhanced MR imaging with histologic tumor grade: comparison of macromolecular and small-molecular contrast media,” AJR Am. J. Roentgenol. 171, 941–9. (1998). [PubMed]
26. M. Y. Su, Z. Wang, P. M. Carpenter, X. Lao, A. Muhler, and O. Nalcioglu, “Characterization of N-ethyl-N-nitrosourea-induced malignant and benign breast tumors in rats by using three MR contrast agents,” J. Magn. Reson. 9, 177–86 (1999). [CrossRef]