1. Introduction

For patients with brain cancer, the extent of tumor resection during neurosurgery is one of the most important factors associated with prolonged survival [1–6]. However, achieving complete tumor resection remains a significant clinical challenge due, in part, to a lack of intraoperative tools available to help differentiate between cancerous and healthy tissue [7–10]. This limitation has devastating consequences for patients, including high recurrence rates (e.g., ∼70% within six months of treatment for high-grade gliomas (HGG) [11]) and exceedingly low survival-times (e.g., <15 months for HGG [11]). It is also important to highlight that unlike other surgical interventions used to treat other types of cancers [12], in brain tumor surgery, biopsies are not taken around the tumor bulk to confirm clear margins. This common procedure is omitted to preserve as much vital, normal brain tissue as possible. Thus, in-vivo assessment is critical, which is far more onerous than ex-vivo analysis (e.g., using frozen section) and further exacerbates the challenges associated with achieving complete resections in brain tumor surgeries.

FDA-approved technologies aimed at addressing this clinical challenge, including neuronavigation and intraoperative CT/MRI, have shown promising results but still have significant limitations. Specifically, neuronavigation is hampered by large brain shifts (>1cm) and positioning errors [4,5]. And intraoperative CT/MRI is (i) prohibitively expensive for many health institutions (>$10 million) [6], (ii) time-consuming, (iii) subject to image artifacts [13], and (iv) unable to provide continuous, real-time intraoperative guidance [14,15]. Unfortunately, neither of these technologies have improved overall survival [4,5,14,15].

More recently, optical methods—which are particularly appealing due to their high resolution, ease of use and low-cost—have been applied to guide neurosurgery. A noteworthy example is fluorescence-guided surgery with 5-aminolevulinic acid (5-ALA): clinical trials using 5-ALA have shown an increase in the extent of resection and improved overall survival for HGG [16–18]. However, the need for exogenous contrast with 5-ALA poses important limitations. For instance, the contrast agent shows variable uptake based on blood-brain barrier permeability, edema, cellular/vascular proliferation, and cancer grade [17]. Administration of the contrast agent prior to surgery must also be adequately timed for optimal results [18]. There is also a lack of information from unlabeled surrounding structures, which is critical for guiding resections in this delicate organ. Finally, 5-ALA has low sensitivity for identifying infiltrative disease and low-grade gliomas [19–21].

In this work we demonstrate that an emerging optical imaging method, quantitative oblique back-illumination microscopy (qOBM) [22,23], clearly identifies normal brain, and both bulk and infiltrative tumor without labels or stains. This technology derives its contrast from refractive index fluctuations encoded in optical phase and provides endogenous information that was—until recently—only accessible in thin, transparent samples using quantitative phase imaging (QPI) [24]. Importantly, qOBM delivers unique 3D (i.e., tomographic), widefield imaging capabilities in thick tissues with clear contrast to cells and their internal contents in real-time, using a simple, flexible, and low-cost imaging system. Using a bulk and infiltrative brain tumor animal model, we show that quantitative features from qOBM provide robust surrogate biomarkers of disease. We also demonstrate qOBM imaging through a compact and flexible, fiber-based system. This novel imaging tool has the potential to identify brain tumor margins intraoperatively and in-vivo which can ultimately help improve surgical outcomes and overall survival for patients with brain tumors. Lastly, due to the versatility of this tool, the same approach can be applied to many other surgical interventions and broader biomedical applications.

2. Materials and methods

2.1 Free-space qOBM system

The The qOBM system uses a conventional brightfield microscope geometry with a few modifications [22,23,25,26]. Instead of a transmission-based illumination, samples are illuminated sequentially in epi-mode with four external, 720 nm LED light sources. The wavelength was chosen to minimize the absorption of light by the tissue, while maximizing the camera’s quantum efficiency. The LEDs are coupled into multimode fibers (1 mm in diameter), arranged 90-degrees from one another around the microscope objective, as shown in Fig. 1(a). When light from an LED source is deployed through one of the fibers into the thick sample (∼45 mW are incident on the sample), light undergoes multiple scattering, causing some of the photons to change trajectory and effectively produce a virtual light source within the thick object, emulating a transmission source with a slight offset to the optical axis. This is known as oblique back-illumination [27]. Under these circumstances, lateral variations in index of refraction at the focal plane refract light toward or away from the acceptance angles of the microscope objective (given by its numerical aperture), producing phase contrast in observed intensity. Images acquired with a pair of diametrically opposed fibers can be subtracted from one another to generate differential phase contrast [27]. Further, due to the large angular extent of the effective illumination (i.e., low spatial coherence of the effective source), and the subtraction process to generate differential phase contrast, out-of-focus content is rejected, permitting tomographic sectioning [27,28]. The penetration depth of the system is of about one free scattering path length, equivalent to ∼100 µm in brain tissue. Further details of the system components have been described elsewhere [22,23,25,26] and are summarized in the Supplement 1.

 

Fig. 1. Schematic of free-space qOBM system and image acquisition. (a) System schematic consisting of inverted brightfield microscope with epi-illumination form 4 LEDs (720 nm) deployed through multimode fibers. The four fibers are spaced 90-degrees from one another using a custom-made 3D-printed fiber-holder adapter. Inset shows a picture of a freshly excised rat brain (9L gliosarcoma rat tumor model) with a coronal cut to expose the tumor (arrow points to small tumor). (b) Raw intensity image from one illumination LED light source in a healthy gray matter brain region. (c) Difference image (OBM) from two diametrically opposed light sources of same region. (d) Processed qOBM image via deconvolution of four intensity images of the same region as (b) and (c). Scale bar = 50 µm.

Download Full Size | PDF

In qOBM, two differential phase images from orthogonal angles (produced from the four light sources) are processed to generate quantitative phase contrast. To quantify the data, we numerically model the angular distribution of the multiple-scattered light passing through the focal plane within the sample, which again, serves as the effective, transmission-like illumination light source [22,23]. This distribution leads to a robust estimate of the optical transfer function (OTF) of the system, which in turn can be used to extract quantitative phase information via a regularized deconvolution (see the Supplement 1 section for details) [22,23,28,29]. The imaging speed of qOBM is dictated by the frame rate of the camera, which we set to 20 Hz, resulting in a net widefield imaging rate of 5Hz (or equivalently, imaging speed of 200 msec.). Processing is performed in real-time.

The simplicity and power of qOBM to reveal detailed cellular and subcellular structures from seemingly uninterpretable brightfield images is illustrated in Figs. 1(b)–1(d). Figure 1(b) shows an example raw image from a single capture with illumination from one LED. Unsurprisingly, this image does not reveal any clear subcellular or cellular detail due to large contributions from out-of-focus planes. More detail is indeed available in the difference image from two diametrically opposed light sources [Fig. 1(c)]. Here out-of-focus contributions are removed but information is only qualitative and generated along one direction with a shadow-like effect (i.e., differential contrast). Finally, after deconvolving the four brightfield images with the system’s unique point spread function (see the Supplement 1 section), the qOBM image [Fig. 1(d)] exhibits remarkable contrast to cellular and subcellular detail, including neurons, axons, and glial cells embedded in neuropil. Importantly, the information in the image is quantitative and can thus be readily applied towards image classification and ultimately rapid intraoperative image-guidance.

2.2 Imaging 9L gliosarcoma rat tumor model with qOBM

To demonstrate the ability of qOBM to identify brain tumors and their margins, we use a 9L gliosarcoma rat tumor model which recapitulates human high-grade gliomas, including high proliferative capability, vascularization and infiltrative pattern (see the Supplement 1 section for more details) [30,31]. All animal experimental protocols were approved by Institutional Animal Care and Use Committee (IACUC) of the Georgia Institute of Technology and Emory University. In short, 14 Fisher rats were implanted 9L gliosarcoma cells intracranially; plus, two additional untreated rats were used as control. For the treated group, tumors were allowed to develop for 9 to 12 days, during which time the tumors grew between 1-10 mm. Next, animals were sacrificed, and brains were excised. A coronal cut was made to expose the tumor, emulating the access surgeons have to tumors during surgery. The inset of Fig. 1(a) shows an excised brain after a coronal cut with a small tumor indicated by the black arrow. Note that the tumor is confined to one side of the brain, leaving the counter-lateral side of all treated brains as an additional control. Without additional processing, fresh brains were placed on the qOBM system and imaged. Then, brains were formalin fixed, embedded in paraffin wax, cut into thin (5 µm) sections and stained with hematoxylin and eosin (H&E) for comparison.

3. Results

3.1 Imaging 9L gliosarcoma rat tumor model with qOBM

qOBM images of fresh, thick brain tissues show marked agreement with the gold standard H&E histopathology images (taken from similar regions). For example, Fig. 2(a) shows a normal grey matter region in which both H&E and qOBM show neurons, as identified by their relatively abundant cytoplasm. These neurons exhibit a rounded morphology and are relatively randomly dispersed in this example region of a deep grey matter structure. Glial cells (mostly oligodendroglial cells) are also clearly visible in both the qOBM and H&E images, although they are slightly more conspicuous in the H&E image because of artifactual perinuclear halos (“fried egg” appearance) caused by immersion fixation [32]. The H&E image in Fig. 2(a) also shows a small capillary, while the qOBM image has a larger vessel with red blood cells still inside. In addition, the neuropil (the network of interwoven neural and glial processes), which is the soft pink (eosinophilic) background in the H&E image, is clearly observed in the qOBM image, with arguably better clarity than H&E for individual fibers. This is likely because qOBM captures the tissue structures in-situ (can also be done in-vivo) without fixing tissues which is known to alter structure [33]. The brighter filament structures in qOBM [bottom right-hand corner in Fig. 2(a)] correspond to a white matter bundle. Figure 2(b) also has multiple white matter bundles, but in this case, the tomographic image captures the tracts as if they were coming out of the page. The corresponding H&E image shows the bundles in a slightly darker pink tone (hypereosinophilic).

 

Fig. 2. qOBM images of freshly excised thick brains samples along with H&E images of similar regions. (a) Normal grey matter. (b) Normal white matter bundles. (c-d) Tumor regions. (e) Well-defined tumor margin. (f) Irregular tumor margin with infiltrating tumor cells. All images are taken with a 60X, 0.7NA objective. Images are acquired ∼50 µm into the tissue. Scale bar = 50 µm.

Download Full Size | PDF

Tumor structures, including solid tumor, tumor margins, and infiltrative tumor regions are also clearly observed in qOBM and are in excellent agreement with H&E. Figures 2(c)–2(d) show two densely hypercellular tumor regions with distinct structures: one [Fig. 2(c)] shows plump, cytologically atypical spindled cells, and the other [Fig. 2(d)] demonstrates similarly malignant-appearing cells with a more epithelioid morphology. Both the H&E and qOBM images show similar overall morphology of the tumor cells in terms of their shape and internal structures. Figures 2(e)–2(f) show tumor margin regions, with Fig. 2(e) showing a well-defined tumor margin, and Fig. 2(f) demonstrating an irregular margin reflective of a degree of tumoral infiltration into adjacent normal neuropil. Note that these infiltrative regions are undetectable with current intraoperative technologies, but are readily detectable with qOBM. Further, the agreement between qOBM (in-situ without tissue processing) and H&E, the gold standard for diagnosis, is exceptional and underscores the potential of qOBM to improve tumor margin assessment in neurosurgery.

3.2 Image classification: towards automated tumor margin assessment

Next, we leverage the quantitative capabilities of qOBM to train and test an image classifier and assess the method’s ability to identify healthy tissue, bulk tumor and tumor margins in a robust and automated fashion. For this task we use data from a 60X, 0.7NA objective and a 20X, 0.45NA objective, independently. (This mirrors the available choices for common, commercially available compact microscope objectives, critical for developing a handheld device for in-vivo operation in the future—more on this below.)

Images from the 20X objective (with a 720µm X 720µm field of view, FoV) and 60X objective (240µm X 240µm FoV) were analyzed at various length scales; that is, images were subdivided into smaller regions of interest (ROIs) (specifically, 240µm X 240µm, 120µm X 120µm, 60µm X 60µm, 30µm X 30µm). Then, independent classifiers were trained and tested for each ROI size. This was done to assess the smallest region necessary to achieve robust classification (i.e., assess the classifier resolution). Then, 482 features per region were extracted (for each length scale and objective) based on texture analysis [34], fractal analysis [35], Fourier space features [36], and mathematical autocorrelation transformations [37,38] (see Supplement 1 for details). Regions were labeled as either healthy or tumor with guidance from an expert neuropathologist. It is important to highlight that in this initial training and testing stage, all tumor margin regions were omitted. To reduce the feature-space dimensionality, feature selection was performed using neighborhood component analysis, solved by stochastic gradient descent [39]. A Gaussian support vector machine (SVM) was trained using regions from 13 animals (11 treated and 2 untreated), and testing was performed on the remaining 3 animals (treated).

Results from the quantitative analysis, summarized in Fig. 3 and Table S1, show that for large ROIs, the classification is near perfect with a near-unity accuracy and area under the curve (AUC) of the ROC analysis. As the ROIs decrease in size, the AUC and accuracy begin to drop off, with the 20X data showing a more precipitous decline than the 60X data. The decrease in AUC and accuracy with decreasing ROI size is expected since larger regions have more information content available to discern between normal or cancerous tissue. However, larger ROIs will have lower sensitivity to small cancerous regions within an otherwise healthy region, and vice versa, leading to poorer resolution for identifying tumor margins and infiltrative disease compared to smaller ROIs. Indeed, as shown in Figs. 3(a)–3(b), the data from 60X objective show a better AUC and sensitivity at lower ROIs but show little to no advantage over the 20X data when analyzed at ROIs > 120µm; moreover, the 20X images have a ∼10-fold larger FoV which is important to enable faster, wide area surveillance. This analysis indicates that both the 120µm X 120µm and 240µm X 240µm ROIs provide a good balance in this trade-off between resolution of classifier and accuracy for identifying tumors, and that the 20X objective (with the larger FoV and ∼1µm resolution) can be used with high accuracy [see Figs. 3(a)–3(b)]. We proceed using an ROI size of 240µm X 240µm since it shows a slightly better performance compared to the 120µm x120µm ROI size with the 20X objective (∼3% greater accuracy and AUC, see Table 1S). We also note that this area is similar to the FoV achieved by our fiber-based qOBM system (∼170µm, more on this below).

 

Fig. 3. qOBM Image classification. (a-b) Accuracy and AUC from an ROC analysis of healthy vs. tumor tissue at multiple ROI sizes using a 20X and 60X objective, respectively. Blue lines correspond to accuracy in the training data set (dash line) and test data set (solid line). The red line corresponds to AUC training data set (dash line) and test data set (solid line). (c) SVM score box plot of all classified data acquired with the 20X objective with ROI size of 240µmX 240 µm. Tumor margin refers to any region containing a mixture of healthy tissue and tumor, or diffused tumor. N gives the number of images per group. (d) Representative images from each group. The marks (circle, star, and triangle) show their specific SVM score and location on the boxplots in (c). (e) SVM score box plot of all classified data acquired with the 60X objective with ROI size 240µmX 240 µm. (f) Representative images from each region. Asterisk give level of significance with p-val≤0.001***, p-val≤0.0001****. All p-values are adjusted for the number of features (Bonferroni correction).

Download Full Size | PDF

We also find that the most important features for differentiating between healthy and tumor tissue include fractal dimension of the phase image and its autocorrelation, phase image coarseness, mean phase value and the highest phase value density (see Supplement 1). This is in agreement with expectations: The observed lower fractal dimension in tumors is indicative of a loss of tissue organization, which is a hallmark of cancer [40–42]. The fractal dimension of the autocorrelation is also lower, indicating a rougher structure of the autocorrelation (Fig. S2), which points to the known increase in structural heterogeneity in tumors [42]. This heterogeneity also leads us to expect coarser texture from tumor images which is indeed observed. Further, the phase value, which is proportional to the refractive index and thus dry mass [43], is known to be higher in cancer cells [44], which is also in agreement with our observations [see Supplement 1 Fig. S3 and Fig. 4(e)]. This can be attributed, in part, to higher chromatin production in cancer cells [45].

 

Fig. 4. Visualization of tumor margins of freshly excised thick brain samples with qOBM. (a) qOBM image of tumor margin region with tumor cells spreading through a blood vessel. (b) Same region with color map indicating SVM score using a sliding window (ROI size) of 240 µm x 240 µm with 20 µm step sizes. (c, d) Well-delineated tumor margin region. (e) Stitched qOBM images (21X31 images, ∼2.3 cm x 1.5 cm) after low-pass filtering to highlight mesoscale structure with phase contrast. The inset shows the H&E image of the same brain after the tissue was fixed, sliced, and stained (all other images are of freshly excised thick brain samples). (f, g, h, i) qOBM images (without low-pass filtering) of select regions in (e) of infiltrative tumor margin, bulk tumor and healthy tissue. (j, k, l, m, n) Corresponding colorized images based on SVM score. Inset in (e) shows an H&E image of the same tumor after processing.

Download Full Size | PDF

Next, the SVM classifier is tested on images containing a mixture of tumor and healthy tissue, and regions with diffused disease (i.e., tumor margin regions). Again, the classifier was not trained on any tumor margin regions. The boxplots in Figs. 3(c) and 3(e) include the entire classification data set (individual training and test box plots are shown in Fig. S4). Remarkably, tumor margins, both well-defined and diffused, have an SVM score that falls between that of the healthy tissue group and tumor group [Figs. 3(c) and 3(e)]. Further, all groups show highly statistically significant differences (all p-values are adjusted for the number of features, i.e., Bonferroni correction). We note that the SVM score distribution of the tumor margin group is wider than the healthy tissue and bulk tumor groups. This is because the amount of tumor present in each margin region varies from a small fraction (e.g., orange circle regions) to nearly taking up the entire ROI (e.g., orange triangles). Finally, Figs. 3(d) and 3(f) show example images from representative regions of each group.

To better visualize tumor margins with qOBM, we produce colormaps based on SVM scores using a sliding window (240µmX240µm) with ∼20 µm step sizes (50 pixels in the 20X images). Images presented in Fig. 4 were not included in the SVM training set. Figures 4(a) and 4(b) show an example with tumor cells infiltrating into normal brain via a blood vessel. Here the classifier can clearly identify the well-defined tumor margin in the bottom left-hand corner, as well as the infiltrating tumor penetrating though a narrow vessel on the right. Moreover, Figs. 4(c) and 4(d) show that the classifier clearly outlines the interface between healthy brain tissue and tumor in a well-defined tumor margin region. Finally, Figs. 4(e)–4(n) show a large brain region consisting of 21 × 31 stitched, 20X qOBM images covering an area of ∼2.3cm x 1.5cm. The grayscale image [Fig. 4(e)] is a low-pass filtered image of the phase values (details in Supplement 1, under Fig. S1), which correlates well with the overall tumor regions, with it showing an overall higher phase value as discussed above. Nevertheless, the colorized images based on SVM scores [Fig. 4(j)–4(n)] provide a more definitive delineation of the tumor regions. Moreover, the classifier is in excellent agreement with manually curated bulk and infiltrative tumor regions (dashed light-blue line and dashed magenta line, respectively in Figs. 4(e) and 4(j)], as determined independently using high-resolution phase images (i.e., without low-pass filtering, see Supplement 1 Fig. S1). We note that there are some false positives around the edges of the tissue, but this is to be expected given that the classifier was not trained on any edge regions. While this could be improved with additional samples and more training, such end-tissue regions are not expected in-vivo. Finally, Figs. 4(f)–4(i) and 4(k)-4(n) show select magnified regions (qOBM images without low-pass filtering) around the tumor margin, including two diffused tumor regions shown in yellow squares [Figs. 4(h) and 4(m)] and cyan squares [Figs. 4(f) and 4(k)]. Magnified regions from bulk tumor [magenta squares, Figs. 4(g) and 4(l)] and a representative healthy region [green squares, Figs. 4(i) and 4(n)] are also presented. In each case, the colorized images show a strong red and green hue in bulk tumor and healthy brain tissue, respectively, while regions containing diffused tumor show a softer red-to-yellow color.

3.3 Fiber based qOBM system: towards intraoperative in-vivo imaging

To demonstrate the translational capabilities of qOBM as a potential intraoperative, in-vivo imaging tool, we develop a fiber-based qOBM imaging system. Our probe design is adapted from existing strategies for fluorescence microendoscopy for in-situ cellular imaging [46]. In short, the fiber-based qOBM system applies an imaging fiber bundle with 30,000 fiber elements, where each individual element is ∼4 µm in diameter (Fujikora). The bundle is attached to a gradient refractive index (GRIN) objective lens (1.4mm diameter and 8mm length) with 0.5 NA. The focal length is fixed at 60 µm (in water) and the magnification is ∼2.6X, resulting in an overall resolution of ∼1.5 µm, dictated by the projection of the fiber elements onto the sample. This resolution is comparable to the resolution achieved with the 0.45 NA, 20X microscope objective used above (∼1µm). As before, sample illumination is in epi-mode, with light delivered sequentially through 1mm fibers around the GRIN lens and fiber bundle [ Fig. 5(a)], all housed in a custom-made 3D-printed enclosure. With this configuration, the distal end of the probe [side closest to sample, Figs. 5(b) and 5(c)] is 4mm in diameter, extremely light, flexible, and suitable for guiding brain surgery. On the proximal end (i.e., the side of the probe closest to the camera), the image relayed through the fiber bundle is detected using a conventional microscope geometry [Fig. 5(d)]. With this configuration, we achieve widefield imaging with a ∼170µm in diameter FoV. The system was validated using 10 µm polystyrene beads immersed in water where all structures are clearly resolvable with good contrast and quantitative agreement (see Supplement 1 Fig. S5). The sensitivity of the probe, assessed by the standard deviation of a background region, is ∼70 nm, which is about an order of magnitude worse compared to the free space system [22], but can be improved in the future with further optimization and using background subtraction, for example.

 

Fig. 5. Fiber-based qOBM system and imaging. (a) Schematic of fiber-based system, along with photographs of (b, c) the distal end and (d) proximal end of probe. (e, f) qOBM images of normal (choroid plexus) and tumor regions, respectively, using fiber-based probe. (g, h) Images of similar brain train structures using free-space qOBM system, and (i, j) corresponding H&E images.

Download Full Size | PDF

Finally, we demonstrate the capabilities of the fiber-based qOBM system to visualize cellular structures in thick brain tissue samples in-situ. For this particular set of experiments, the thick brain tissues were formalin-fixed. Figures 5(e)–5(j) show a healthy choroid plexus region and a tumor region as detected with both the fiber-based and free-space qOBM systems, along with corresponding H&E image after tissue processing. In both of these examples, the fiber-based qOBM system clearly shows cellular structures and is in excellent agreement with the free-space system and the H&E images, further demonstrating the potential of qOBM to be applied in-vivo to guide brain tumor resection.

4. Conclusion and discussion

In this work, we have shown that qOBM clearly identifies brain tumor margins in a 9L gliosarcoma rat tumor model which recapitulates human high-grade gliomas. The structural detail provided by qOBM, based on quantitative phase contrast, has remarkable agreement with the gold standard H&E stained tissue sections; however, qOBM can provide information in real-time, without exogenous labels in-situ and potentially in-vivo. Due to the quantitative capabilities of this technology, robust surrogate biomarkers of disease based on phase values and structure can be extracted and applied for automated tumor and margin detection and visualization. Finally, we have developed a fiber-based qOBM system that is apt for in-vivo intraoperative use. Moreover, we demonstrated the fiber-based system’s ability to yield similar structural detail as the free-space qOBM system. Taken together—the clear cellular and subcellular detail, robust automated classification/tumor detection, and flexible, low-cost, handheld embodiment of the fiber-based qOBM system—this work has significant implications for label-free, in-vivo brain tumor margin detection, which can help address an important unmet clinical need.

Recently, there have been other important developments in label-free optical imaging applied to brain tumor margin detection. Stimulated Raman scattering (SRS), for example, has been used to image excised brain tissue samples, where SRS images can not only be processed to appear nearly identical to H&E, but can also be used for near real-time diagnosis [47,48]. However, SRS is a nonlinear technology and thus requires a complex and expensive optical system. More importantly, for neurosurgical applications, SRS has only been demonstrated using excised tissues; and while encouraging, substantial and nontrivial technological hurdles need to be overcome to enable in-vivo intraoperative tumor margin assessment. Optical coherence tomography (OCT) [49] and spontaneous Raman scattering [50] have both been applied to detect tumor margins intraoperatively and in-vivo, however these technologies lack critical cellular and subcellular information (available to both SRS and qOBM). In contrast, qOBM offers new and highly desirable capabilities for neurosurgery, including quantitative cellular and subcellular information in real-time using a low-cost and simple optical system that can be applied in-vivo. Moreover, qOBM can potentially be integrated with other optical methods, including fluorescence, spontaneous Raman and OCT, to provide complementary information in-vivo.

In summary, qOBM enables quantitative phase contrast in thick scattering samples, overcoming a significant limitation of QPI that had previously hindered its use in clinical and other translational applications. In fact, to the best of our knowledge, this work presents the first demonstration of quantitative phase contrast for a clinical or surgical application. The capabilities of qOBM make it possible to visualize cellular and subcellular structures using a simple and low-cost configuration that is suitable for imaging both excised and/or in-vivo tissue samples for intraoperative image guidance and more. The level of detail provided by qOBM can not only be used to guide surgery, but can also be used to better assess prognosis, given that qOBM may help visualize the degree of inoperable diffused tumor remaining in a patient. Thus, we expect qOBM to become an important tool for guiding brain tumor surgery. Furthermore, given its unique capabilities, qOBM can be widely applicable in other biomedical and clinical areas.