Quantitative phase imaging to improve diagnostic accuracy of urine cytology

Abstract

BACKGROUND

Definitive diagnosis of urothelial carcinoma in urine cytology is often challenging and subjective. Several urine cytology samples receive an “indeterminate” diagnosis. Ancillary techniques such as fluorescence in situ hybridization (FISH) have been employed to try improve diagnostic sensitivity, but this is not approved as a routine screening test and the complex fluorescent staining protocol also limits its widespread clinical use. Quantitative phase imaging (QPI) is an emerging technology allowing accurate measurement of single-cell dry mass. This study was undertaken to explore the ability of quantitative phase imaging to improve the diagnostic accuracy of malignancy in urine cytology.

METHODS

We performed quantitative phase imaging on unstained ThinPrep-prepared urine cytology slides from 28 patients with four categories of cytological diagnosis (negative, atypical, suspicious and positive for malignancy). Nuclear/cell dry mass, their entropy and nucleus-to-cell mass ratio were calculated for several hundred cells for each patient, and were then correlated with follow-up diagnoses.

RESULTS

The nuclear mass and its entropy of urothelial cells showed significant difference between negative and positive groups. These data showed progressive increase from patients with negative to atypical/suspicious and positive cytologic diagnoses. Most importantly, for those patients in atypical or suspicious groups, the nuclear mass and its entropy from those patients with a follow-up diagnosis of malignancy were significantly higher than those patients without a subsequent malignant follow-up.

CONCLUSIONS

Quantitative phase imaging shows potential to improve the diagnostic accuracy of urine cytology, especially for indeterminate cases and should be further evaluated as an ancillary test for urine cytology.

INTRODUCTION

Urine examination is one of the oldest clinical laboratory tests which was first popularized by George Papanicolaou and Marshall in the 1940s,¹ and remains an essential screening tool in high-risk situations and as a surveillance modality for the detection of urothelial neoplasia.^2–4 The accuracy of urine cytology diagnosis depends on several factors,⁵ such as type of specimen⁶, sample adequacy^7–9 and processing method, underlying clinical condition, tumor grade, and sometimes interpreter variability¹⁰. In practice, its diagnostic accuracy can be confounded by poorly preserved cells, inflammation, infection and low numbers of cytologically atypical cells, making it difficult even for expert cytopathologists to render a definitive diagnosis. This equivocal diagnostic category of “atypical urothelial cells” has caused significant frustration for cytopathologists and clinicians.^11,12 It was reported that approximately 30–35% of urine cytology samples receive an “indeterminate” diagnosis,⁶ with a sensitivity and specificity for diagnosing low-grade tumors to be as low as of 8.5% and 50%, respectively⁶. Furthermore, given that the diagnostic criteria for this indeterminate category are often subjective and used variably by individual cytopathologists, there is significant interobserver variability.¹³ Thus, in the absence of cytopathologist agreement and the lack of non-overlapping standardized diagnostic criteria for urothelial atypia, the atypical category remains a challenge.

Ancillary techniques such as fluorescence in situ hybridization (FISH) have been employed to improve the sensitivity of urine cytology. Multiple chromosomes, such as 1, 3, 4, 7, 8, 9, 11, and 17, are altered in urothelial tumors.¹⁴ A FISH assay can usually detect these chromosomal alterations. UroVysion™ (Vysis, Downers Grove, IL) is such a multi-target FISH assay, approved by the FDA for both monitoring of patients with a history of bladder cancer and detection in patients with hematuria. Despite the fact that FISH outperforms urine cytology across all stages and grades of urothelial carcinoma as reported in several comparative studies,^15,16 there are limitations to this test. It involves a complex fluorescent staining protocol and specialized fluorescent imaging instrument, and specially trained technologists to distinguish subtle differences in banding patterns on bent and twisted metaphase chromosomes; thus, FISH testing can be of high cost and not necessarily suitable for a point-of-care test, which limits its clinical usefulness. Furthermore, despite its high sensitivity, FISH requires an adequate number of exfoliated abnormal cells to be present in a urine specimen. Thus, small urine volume, with low numbers of abnormal cells may preclude the fulfillment of positivity criteria.

Quantitative phase imaging (QPI) is an emerging technology in which quantitative phase images of unstained specimens can be obtained.¹⁷ In the early 1950s, it was recognized that the optical phase shift induced by biological cells of interest is linearly proportional to the dry mass (non-aqueous content) of the cell. Since then, QPI has been widely used to monitor cell dry mass.^18–22 Recently, it has been shown, both theoretically and experimentally, that the surface integral of the cell phase map is invariant to small osmotic changes,²³ which further supports QPI methods to be used for dry mass measurements.

In this proof-of-concept study, we evaluated the use of QPI-based on diffraction phase microscopy (DPM) to quantify cell and nuclear dry mass of urothelial cells from unstained, ThinPrep-prepared slides of urine samples. The aim was to investigate the potential of this imaging modality to be used as a complementary tool to aid cytopathologists in making objective and precise diagnoses, especially for those patients in “indeterminate” groups.

MATERIALS AND METHODS

Sample collection and preparation

All studies were approved by the University of Pittsburgh Institutional Review Board. Urine specimens were collected from 28 patients using different methods (voided or with instrumentation) as part of the patients’ standard clinical care at a University of Pittsburgh Medical Center (UPMC) hospital. The cases were selected to represent four categories of cytological diagnosis (negative, atypical, suspicious and positive for malignancy), which also have sufficient number of urothelial cells present. Urine samples were submitted fixed in Cytoloyt^® from which liquid-based Thinprep^® cytology slides (Hologic, Boxborough, MA) were prepared. An extra unstained Thinprep^® slide was prepared from residual vial material for each case and coverslipped using Cytoseal mounting medium (Richard-Allan Scientific). Urothelial cells from four categories of cytological diagnosis (negative, atypical, suspicious, and positive for malignancy) were imaged with a custom-built quantitative phase microscope. We imaged all urothelial cells, not just cytologically abnormal appearing ones. Among these 28 patients, 21 patients had follow-up histopathology (15 patients) or urine cytology. No patient with negative diagnosis had follow-up histopathology and some of them did not have follow-up urine cytology, which we assumed true negative cases due to the fact the low likelihood of malignancy compared to other cytologic diagnostic categories. The clinical cytology diagnosis and follow-up results were then correlated with our QPI results.

Diffraction Phase Microscopy imaging system

We custom-built a QPI system in transmission mode known as a diffraction phase microscope,^24,25 as illustrated in Fig. 1. We used a low-coherence, laser-driven white-light source (Energetiq EQ-99X) for illumination. Three off-axis parabolic mirrors were used to collimate and de-magnify the divergent output light beam of the lamp. The collimated white-light beam was then fed into an acousto-optic tunable filter (AOTF) for wavelength-tuning in the visible range with a spectral resolution of 1–3 nm. The collimated, single-color light was then coupled into a standard commercial microscope (Observer A1, Zeiss). Using a broad-spectrum light source and an AOTF, we can measure phase images at different wavelengths with high spatial and temporal sensitivities. For this study, only a single wavelength (565 nm) was used. We used a Zeiss LD Plan-NEOFLUAR 20×/0.4 NA objective for image acquisition. A diffraction phase microscopy (DPM) module was added to the side port of the microscope for QPI. Specifically, a transmission grating with 110 grooves/mm was placed at the image plane to generate multiple copies of the emerging field. The 4f-lens system was used to generate off-axis interference on the camera. At the back focal plane of the lens L1, the zeroth-diffraction order was filtered using a 50µm pinhole to pass through only the DC component of the field, while the entire first-order diffracted beam was also passed through; all other diffraction orders were blocked. After passing through the lens L2, the two beams converged and interfered on the camera to create an interferogram that has the following irradiance distribution:

$$I(x,y)=I_0+I_1(x,y)+2\sqrt{I_0{I}_1(x,y)}\mathit{\text{cos}}[kx+\phi (x,y)]$$ (1)

Where I₀ is the irradiance of the zeroth diffraction order after passing the pinhole, I₁(x, y) is the irradiance of the first diffraction order, k is the spatial modulation frequency provided by the grating, and φ(x, y) is the phase shift induced by the object of interest. The captured interference at each wavelength is then used to reconstruct the phase image as previously described in the literature.^24,26 Specifically, the sinusoidal term in Eq. (1) can thus be isolated by Fourier high-pass filtering. The filtered Fourier spectrum was then shifted to center at zero Fourier frequency to remove the modulation frequency k. Finally, the phase induced by the object, φ(x, y), was extracted using a simple inverse tangent operation.

Figure 1.

Schematic setup of diffraction phase microscopy system. OAP: Off-Axis Parabolic Mirror; AOTF: Acousto-Optics Tunable Filter; M: Mirror; TL: Tube lens; Grating: Transmission diffraction grating; L1, L2: Lens; EFL: effective focal length; FL: focal length

For data acquisition, the unstained slide was automatically scanned to acquire interferograms throughout the entire slide (15×15mm²). To minimize the out-of-focus effect, we first automatically scanned small 6×6 field of views (FOVs), with each FOV covering an area of 250×300 µm² of the slide, and then after each 6×6 scan, we refocused the image (done manually or automatically). After imaging the entire unstained slide, the slide was placed in xylene to remove the coverslip and subsequently stained using the Papanicolaou staining protocol. The stained slide was then imaged using the same procedure as that for the unstained slide. The images of stained cells were used to facilitate subsequent nuclear segmentation and cytopathology correlation.

Cell Segmentation

Automatic segmentation of cells and nuclei was performed on quantitative phase images of unstained slides for most urothelial cells. However, for those cases where automatic segmentation did not work properly, stained images of the samples were acquired and registered with corresponding unstained images for nuclei segmentation. For automatic segmentation of urothelial cells, we used simple threshold, erosion, and dilation operations to generate binary masks. Upper and lower limits on sizes of cells were used to eliminate squamous cells and blood cells. We then used an automatic connected component labeling function to assign different labels to different cells. Automatic nuclear segmentation was performed for each cell using adaptive thresholding. However, because of the large variation in size of urothelial cells and the presence of other cell types and debris in these samples, manual segmentation was also implemented to add cells that failed to get detected, or to remove unwanted cells/debris segmented by the automatic program. The software was implemented in MATLAB (MathWorks).

Dry mass calculation and data analysis

The dry mass density at each pixel was calculated as:

$$\rho (x,y)=\frac{\lambda }{2\pi y}\phi (x,y),$$

where λ is the center wavelength, and γ is the average refractive increment of protein. We used γ = 0.2 mL/g, which corresponds to an average reported value²¹, and φ(x, y) is the measured phase. The total dry mass was then calculated by integrating over the region of interest in the dry mass density map. The 1 nm spatiotemporal sensitivity of our DPM imaging system translates into a sensitivity of 5 fg/µm². The segmentation masks for urothelial cells and corresponding nuclei were mapped onto phase images to calculate a set of physical parameters (cell and nucleus dry mass, nucleus-to-cell dry mass ratio calculated by nucleus mass divided by whole cell dry mass, entropy of nuclear dry mass) for each individual cell. Entropy of nucleus dry mass was calculated as:

$$E_{\rho }=-{\sum }_{i=1}^{N}p({\rho }_i){\text{log}}_{2}p({\rho }_i),$$

where dry mass is considered as a discrete random variable ρ_i, ρ denotes the probability which ranges from ρ₁ to ρ_N. For each patient, we were able to collect from 50 to more than 400 urothelial cells (depending on the cellularity of the urine cytology slide) for data analysis.

Statistical Analysis

The statistical comparison between two groups used throughout the entire study was obtained using Student t-test at 95% confidence interval, and two-sided P-values were used for all analyses (Microsoft Excel 2013). A P-value of 0.05 or less is considered as statistical significance. To obtain the characteristic value for an individual patient, we obtained the mean value of each parameter by taking the average value of ~50–400 cells or cell nuclei from each patient, with a prefix of “patient average” used in front of the name of each parameter.

We also constructed a receiver operating characteristic (ROC) curve to evaluate the performance characteristics for each of the parameters using the univariate logistic regression model (Matlab Statistical Toolbox, Mathworks) to classify malignancy using the follow-up clinical diagnosis as the outcome. The discriminant power of the model was assessed by means of the area under the ROC curve.

RESULTS

We imaged a total of 28 patients from four categories of cytological diagnoses (negative, atypical, suspicious, and positive for malignancy). All diagnoses were made by cytopathologists. Table 1 shows the information of patients including their clinical history, cytology diagnosis, and follow-up diagnosis (based on available histology and/or cytology samples). Figure 2 shows an example of quantitative phase images of unstained urothelial cells and their corresponding Pap-stained images. The quantitative phase images shown provide high image contrast for unstained cells with low noise and a clean background. Cell nuclei can be easily distinguished from the cytoplasm, which is very important for efficient automatic cell and nuclei segmentation. Furthermore, the phase value of the nuclei provides a precise quantitative assessment of nuclear dry mass. For example, in Fig. 2c, the cell nuclei from a patient with a “suspicious for malignancy” cytology sample (patient number 10 who developed high grade urothelial carcinoma confirmed by histology) are brighter (higher nuclear dry mass) compared to the negative case (Fig. 2a), consistent with more condesned chromatin seen in the Pap-stained cytology image. This suggests that nuclear dry mass from unstained cells may serve as a quantitative measure for detecting bladder cancer. Figures 3(a,b) show representative histograms of the dry mass of urothelial cells and their nuclei, respectively, from four patients in each of the four diagnostic categories and N indicates the number of urothelial cells analyzed for the corresponding patient. The nuclear dry mass shows a progressively increased value from patients with cytologic diagnosis of negative to atypical, suspicious, and to positive for malignancy. Figures 3(c,d) show the box plots of dry mass for all urothelial cells and their nuclei from each diagnostic category. An important observation is that the total cell dry mass in Fig. 3(c) does not exhibit any trend among the 4 cytologic diagnostic categories. The average total cell masses of the negative, atypical, suspicious and positive groups are 51.1 pg, 49.2 pg, 59.5 pg and 52.6 pg, respectively. However, the box plots of nuclear dry mass in Fig. 3(d) show a significant difference among these groups with an increasing trend from the negative to positive category. This result suggests that it is the urothelial nuclear dry mass, rather than the total cell dry mass, that correlates with cytologic diagnosis of urothelial malignancy.

Table 1.

Patients’ diagnosis at time of sample collection and follow-up data

Case No	Age (years)	Gender	Collection method	Clinical history	Cytology diagnosis	Histology (or urine) follow-up
1	69	Male	NOS	Prostate cancer	Atypical	None
2	80	Male	NOS	HG urothelial carcinoma	Atypical	Urothelial carcinoma in situ
3	60	Male	Voided	Liver disease	Atypical	None
4	79	Male	Voided	HG urothelial carcinoma	Atypical	Papillary urothelial carcinoma HG non-invasive
5	66	Male	Voided	HG urothelial carcinoma	Atypical	Benign (inflammation)
6	82	Male	NOS	Urine retention	Negative	(Negative)
7	38	Male	NOS	Skin melanoma	Negative	(Atypical)
8	58	Female	NOS	Renal cell carcinoma	Negative	None
9	89	Female	NOS	HG urothelial carcinoma	Negative	(Negative)
10	64	Male	Voided	Atypical urine cytology	Suspicious	HG urothelial carcinoma
11	77	Male	NOS	HG urothelial carcinoma	Positive	HG urothelial carcinoma
12	81	Male	Voided	Hematuria	Suspicious	HG urothelial carcinoma
13	65	Male	NOS	Atypical urine cytology	Suspicious	Urothelial carcinoma in situ
14	78	Female	NOS	Urothelial dysplasia	Positive	Urothelial carcinoma in situ
15	85	Male	Voided	Urothelial carcinoma in situ	Suspicious	(Suspicious)
16	72	Female	Renal pelvis barbotage	Renal mass	Positive	(Positive)
17	73	Male	NOS	HG urothelial carcinoma	Positive	HG urothelial carcinoma
18	76	Male	NOS	Prostate cancer	Positive	Ureter mass (not biopsied)
19	56	Male	Voided	None given	Suspicious	HG urothelial carcinoma
20	59	Male	Voided	None given	Atypical	LG papillary urothelial carcinoma
21	83	Female	Voided	HG urothelial carcinoma	Atypical	HG urothelial carcinoma
22	81	Female	NOS	None given	Atypical	HG urothelial carcinoma
23	61	Male	NOS	None given	Atypical	(Atypical)
24	59	Male	Washing	HG urothelial carcinoma	Positive	HG urothelial carcinoma
25	78	Female	Voided	HG urothelial carcinoma	Negative	None
26	72	Male	Voided	HG urothelial carcinoma	Negative	None
27	51	Female	Voided	Hematuria	Negative	None
28	76	Female	Voided	Breast cancer	Negative	None

Abbreviations: LG = low grade; HG = high grade, NOS = not otherwise specified those with parenthesis only at the last column indicates that only urine cytology follow-up is available.

Figure 2.

Examples of reconstructed phase images of unstained urothelial cells from 4 representative patients in each of the following 4 diagnostic categories: Top row: (a) Negative; (b) Atypical; (c) Suspicious; and (d) Positive for urothelial carcinoma. Color bar is in radian. Bottom row: Corresponding Pap-stained images of the top row.

Figure 3.

Histograms of cell dry mass (a) and nuclear dry mass (b) of urothelial cell populations of 4 representative patients in each of the 4 diagnostic categories. Box plot of cell dry mass (c) and nuclear dry mass (d) of all urothelial cells in four diagnostic categories.

Furthermore, from the quantitative phase images, we can calculate several other morphological parameters of urothelial cells including both conventional parameters used in urine cytology (i.e., nucleus-to-cytoplasm (N/C) ratio, nuclei diameter (i.e., equivalent diameter assuming circular shape of the cell nuclei), cell diameter (size)) and QPI-derived parameters (i.e., nuclear dry mass, nuclear entropy and nucleus-to-cell dry mass ratio (NCMR)). Next to evaluate the performance characteristics of each parameter, Fig. 4 shows receiver operating characteristic (ROC) curves for each of the 6 parameters. The conventional cytologic parameters of mean cell mass and the nuclear size performed poorly with the area under the curve (AUC) of 0.64 and 0.58, respectively. The conventional N/C ratio achieved an AUC of 0.88, slightly worse than that of NCMR. The nuclear dry mass and nuclear entropy had the highest AUC of 0.98 and 1, respectively. The nuclear dry mass quantifies the integrated mass of the entire nuclei and an increased nuclear dry mass may be associated with the higher DNA content; while an increased nuclear entropy suggests a higher structural heterogeneity within the nucleus.

Figure 4.

Receiver operating characteristic (ROC) curves for classifying malignancy using each of the patient average parameters calculated from quantitative phase images.

Figures 5(a,b) show the box plots of the “patient average” total cell and nuclear dry mass, which were calculated by averaging all urothelial cells or cell nuclei for each patient. Again, the patient average total cell dry mass does not show any correlation with cytologic diagnosis, but the average nuclear dry mass shows a progressive increase from negative to positive groups, with large separations between groups (p-value < 0.01). Furthermore, Fig. 5(c,d) shows the scatter plots of the patient average nuclear dry mass entropy versus the patient average total nuclear dry mass and nucleus-to-cell dry mass ratio (NCMR) for each patient, respectively. The two extreme categories of negative and positive are well separated via nuclear dry mass, NCMR, and nuclear entropy. Among the 7 patients with an atypical cytology diagnosis who had follow-up, those who had confirmed neoplasia (carcinoma in situ or carcinoma) (case numbers 2, 4, 20, 21 and 22) are well-separated from those without malignancy (inflammation) (case numbers 5, and 23), as indicated by the red and blue circles, respectively, in Fig. 5(c,d).

Figure 5.

(a) Box plot of patient average cell dry mass; (b) Box plot of patient average nuclear dry mass; (c) Scatter plot of patient average nuclear dry mass and nuclear entropy; (d) Scatter plot of patient average nucleus-to-cell dry mass ratio and nuclear entropy. Green dots in red and blue circles indicate patients with atypical cytology with follow-up histopathology with and without an advanced diagnosis of carcinoma in situ or urothelial carcinoma, respectively.

Among the 4 suspicious cases with follow-up histology diagnosis (case numbers 10, 12, 13 and 19), all had histology confirmed malignancy (carcinoma in situ or invasive urothelial carcinoma). Figures 5(c,d) show that patients with a suspicious cytologic diagnosis were all classified by QPI as positive, with a similar range of nuclear entropy to that of the positive cases. Of note, patient 18 who was initially diagnosed as positive for malignancy on cytology and who then developed metastatic bladder cancer had one of the highest values of nuclear dry mass, nuclear entropy and NCMR.

From the above analyses, we chose the three best parameters, namely, nuclear entropy, nuclear dry mass, and NCMR as the main predictors. We proposed the following multivariate scoring scheme using three calculated parameters: average nuclear mass, average nuclear entropy, and average nuclear to cell mass ratio. Each parameter was tested against a chosen threshold that results in a TRUE or FALSE outcome. The thresholds for nuclear mass entropy (NE) and nuclear mass (NM) were selected according to the optimal operating points on the ROC curves, which achieved false positive and true negative rates of 0 and 1, respectively, for NE and 0 and 0.94, respectively, for NM. The threshold for NCMR of 0.5 was chosen by following the standard cutoff value of N/C ratio of 0.5 commonly used in conventional cytology to identify cancerous cells. Then, a diagnosis for each patient was based on a majority vote of the three test outcomes. Table 2 shows our prediction results together with an initial and subsequent clinical follow-up diagnosis. Our prediction results show a perfect match with the patients’ available follow-up outcomes. Needless to say, this is just a proposed scheme based on preliminary results in a limited number of patients. A large-scale study is needed to verify this scoring scheme as well as to train the test thresholds to validate a rigorous diagnosis scoring scheme. These thresholds and the scoring scheme may be adjusted and new quantitative parameters may also be identified when a larger training data set is available in the future large-scale study.

Table 2.

Proposed scoring scheme for diagnosis.

Case Number	Cytology Diagnosis	Follow-up histology (or urine)	Test			Our Results
			NM > 21pg	NE > 0.9	NCMR > 0.5
1	Atypical	None	1	1	0	1
2	Atypical	Urothelial carcinoma in situ	1	1	0	1
3	Atypical	None	0	0	0	0
4	Atypical	Papillary urothelial carcinoma HG non-invasive	1	1	0	1
5	Atypical	Benign (inflammation)	1	0	0	0
6	Negative	(Negative)	0	0	0	0
7	Negativel	(Atypical)	0	0	0	0
8	Negative	None	0	1	0	0
9	Negative	(Negative)	0	0	0	0
10	Suspicious	HG urothelial carcinoma	1	1	0	1
11	Positive	HG urothelial carcinoma	1	1	1	1
12	Suspicious	HG urothelial carcinoma	1	1	0	1
13	Suspicious	Urothelial carcinoma in situ	1	1	0	1
14	Positive	Urothelial carcinoma in situ	1	1	1	1
15	Suspicious	(Suspicious)	1	1	1	1
16	Positive	(Positive)	1	1	1	1
17	Positive	HG urothelial carcinoma	1	1	1	1
18	Positive	Ureter mass (developed metastastic bladder cancer)	1	1	1	1
19	Suspicious	HG urothelial carcinoma	1	1	1	1
20	Atypical	LG papillary urothelial carcinoma	1	1	0	1
21	Atypical	HG urothelial carcinoma	1	1	0	1
22	Atypical	HG urothelial carcinoma	0	1	1	1
23	Atypical	(Atypical)	0	0	1	0
24	Positive	HG urothelial carcinoma	1	1	1	1
25	Negative	None	0	0	0	0
26	Negative	None	0	0	0	0
27	Negative	None	0	0	0	0
28	Negative	None	0	0	0	0

Abbreviations: NM = Nuclear mass, NE = Nuclear entropy, NCMR = Nuclear to cell mass ratio those with parenthesis only at the 3rd column indicate that only urine cytology follow-up is available.

CONCLUSION AND DISCUSSION

Urine cytology remains the most widely used screening method for the detection of urothelial carcinoma of the urinary bladder, due to its non-invasive nature and low cost, compared to the gold standard of cystoscopy and biopsy. Despite its high sensitivity in detecting high-grade bladder tumors, urinary cytology is much less sensitive in detecting low-grade tumors with a sensitivity of 5–50%.^2,6 The atypical diagnostic category remains a challenge in cytopathology, with inter-observer variability among individual cytopathologists even in the same institution. A significant number of patients who have atypical cytology are not biopsied or have a corresponding lesion detected histologically, even though this category includes both specimens that are reactive or may be attributed to a neoplastic lesion. Ancillary techniques such as FISH have shown superior performance over cytology,^15,16 but due to the high cost, difficulty with establishing proper fluorescence staining and the need for well-trained technologists to interpret FISH results, it is still not widely adopted in many medical centers. An accurate, non-invasive, label-free and low-cost test would be very useful for both screening of high-risk populations and for monitoring patients with a history of bladder cancer to help identify recurrence early and prevent disease progression.

QPI addresses several current limitations in the cytologic diagnosis of urothelial carcinoma. First and foremost, it offers a precise quantitative assessment of nuclear dry mass to assess nuclear features, which addresses inter-observer variability; second, compared to FISH, the QPI test is label free, which significantly simplifies the sample preparation protocol, thus potentially low cost as it eliminates the need for fluorescent probes; third, this imaging system only requires a simple add-on optical module (including a transmission grating and a couple of lenses) to a regular bright-field microscope, and thus is of lower cost compared to fluorescence imaging used in FISH; fourth, the entire imaging process as well as image processing procedure for extracting diagnostic parameters can be fully automated, which can potentially remove the need for a specially trained technologist. Although the eventual cost depends on the clinical adoption, the label-free nature and simple optical instrument are key factors for low-cost device.

Conventional phase microscopy suffers from the well-known halo artifacts, which may negatively affect the sensitivity of the dry-mass based quantification. Such effect is minimized in our DPM system with a laser-driven Xenon light source which provides high intensity and high spatial coherence. However, to further reduce the instrument cost, other broadband light sources may be used, such as light-emitting diode, but the reduced spatial coherence may aggravate the halo artifact, which may be minimized by utilizing post-processing computational approaches to remove halo-artifact,²⁷ or other QPI methods²⁸

In summary, QPI was able to accurately assess urothelial cell nuclear dry mass and its entropy which serve as a quantitative marker for urothelial neoplasia. With our small dataset, these imaging markers show promise to assist with challenging atypical cases by separating patients with a follow-up diagnosis of malignancy form those without. QPI is able to analyze all urothelial cells, not just the abnormal cells. As shown in the histogram of nuclear dry mass (Fig. 3), the overall distribution of nuclear dry mass for the entire cell population, not just a limited number of cytologically abnormal cells, is shifted to higher values similar to cytology specimens from patients with urothelial carcinoma.

Despite these promising results, due to the limited number of patients investigated, future large-scale clinical studies are needed to further validate the specificity and sensitivity of QPI imaging for bladder cancer detection. Furthermore, as discussed above, a large-scale study will help develop a more precise and rigorous multivariate scoring scheme. Also, further studies are required to address potential technical (e.g. sample collection methods) and clinical (e.g. polyomavirus infection) confounding issues that may affect QPI performance. Nevertheless, this study demonstrates that QPI has great potential as a cost-effective, objective and quantitative test to improve diagnostic accuracy for bladder cancer detection. In the future, we envision the potential clinical use of QPI as an ancillary test to enhance the detection of malignancy in urine cytology where conventional methods cannot make a definitive diagnosis. Alternatively, if whole-slide imaging system is widely adopted in cytopathology, this method may also be used as a quantitative automated (or semi-automated) cytologic evaluation. A few technical issues will need to be addressed for a fully-automated system incorporated into the whole-slide imaging system, such as autofocusing, better automatic cell and nuclear segmentation algorithms.