Deep Learning-Based Screening of Urothelial Carcinoma in Whole Slide Images of Liquid-Based Cytology Urine Specimens

Urinary cytology is a useful, essential diagnostic method in routine urological clinical practice. Liquid-based cytology (LBC) for urothelial carcinoma screening is commonly used in the routine clinical cytodiagnosis because of its high cellular yields. Since conventional screening processes by cytoscreeners and cytopathologists using microscopes is limited in terms of human resources, it is important to integrate new deep learning methods that can automatically and rapidly diagnose a large amount of specimens without delay. The goal of this study was to investigate the use of deep learning models for the classification of urine LBC whole-slide images (WSIs) into neoplastic and non-neoplastic (negative). We trained deep learning models using 786 WSIs by transfer learning, fully supervised, and weakly supervised learning approaches. We evaluated the trained models on two test sets, one of which was representative of the clinical distribution of neoplastic cases, with a combined total of 750 WSIs, achieving an area under the curve for diagnosis in the range of 0.984-0.990 by the best model, demonstrating the promising potential use of our model for aiding urine cytodiagnostic processes.

Improved efficiency of urine cell image segmentation using droplet microfluidics technology

Recent advances in the recognition of biological samples using machine vision have made this technology increasingly important in research and detection. Image segmentation is an important step in this process. This study focuses on how to reduce the interference factors such as the overlap between different types (or within the same type) of urine cells according to microfluidics and improve the machine vision segmentation accuracy for cell images. In this study, we demonstrate that the platform can realize this hypothesis using urine cell image segmentation as an example application. We first discuss the reported urine cell droplet microfluidic chip system, which can realize the test conditions in which urine cells are encapsulated in the droplet and isolated from salt crystallization and/or bacteria and other urine-formed elements. Then, based on the analysis conditions set in the aforementioned experiment, the proportions of red blood cells, white blood cells, and squamous epithelial cells covered by various formed elements in the total urine cells in the same urine sample are measured. We simultaneously analyze the percentage of urine cells covered by salt crystallization and the incidence of overlapping between urine cells. Finally, the Otsu algorithm is used to segment the urine cell images encapsulated by the droplet and the urine cell images not encapsulated by the droplet, and the Dice, Jaccard, precision, and recall values are calculated. The results suggest that the method of encapsulating single cells based on droplets can improve the image segmentation effect without optimizing the algorithm.

Diagnostic criteria of sentinel lymph node micrometastasis or macrometastasis based on tissue rinse liquid-based cytology in gynecological cancer

OBJECTIVE

This study aimed to establish intraoperative diagnostic criteria of sentinel lymph node (SLN) micro-/macrometastasis on the basis of tissue rinse liquid-based cytology (TRLBC) in gynecological cancer.

METHODS

We enrolled 214 patients with gynecological cancer who underwent rapid diagnosis of SLN metastasis on the basis of TRLBC from a total of 490 SLNs. For slides that were classified as positive for atypical cells on cytological inspection, we counted the number of clusters (an atypical cell mass consisted of three or more cells) and the number of single cells (an atypical cell other than clusters). Receiver operating characteristic (ROC) analysis was applied to determine the efficiency of predicting SLN micro-/macrometastasis.

RESULTS

On cytological inspection, 36 slides were classified as positive for atypical cells, while 21 slides (4.3%) were true positive, 15 (3.1%) were false positive, and 454 (92.6%) were true negative. There were no false negative results in this study. The area under the ROC curve for the number of cluster was superior to that for the number of single cells for distinguishing micro-/macrometastasis from negative/isolated tumor cells (0.86 vs. 0.67, P = 0.032). The optimum cut-off value of the number of clusters was 5 for distinguishing these two categories.

CONCLUSIONS

TRLBC is a highly sensitive alternative for detecting SLN metastasis as a rapid intraoperative diagnosis. Counting the number of atypical cell clusters might be useful for distinguishing micro-/macrometastasis from isolated tumor cells.

Cytomorphologic features and differential diagnosis of neoplasms with small cell features in liquid-based urinary tract cytologic specimens

INTRODUCTION

Urinary tract cytology (UTCy) allows the accurate diagnosis of high grade urothelial carcinoma. Rare malignancies characterized by the presence of small cells may be more difficult to recognize, however. The aim of this study was to review our experience with liquid-based (ThinPrep) UTCy specimens showing small atypical cells and characterize their cytomorphology, potential differences from previously reported morphologic features, and discuss their differential diagnostic considerations.

MATERIALS AND METHODS

Among 18,859 UTCy specimens reviewed during a 13-year period (2001-2012), we identified 13 cases corresponding to surgical pathology specimens diagnosed as small cell carcinoma (6), melanoma (3), lymphoma (3), and leukemic involvement of the urinary tract (1). We recorded the original diagnoses made on these cases and 10 cytomorphologic features that could aid in their diagnosis.

RESULTS

We identified 13 cases in UTCy of 7 men and 6 women; 11 of which were diagnosed as positive or suspicious for malignancy. In 8 out of 13 cases (62%) the type of malignancy was correctly reported. Of the 10 recorded features, cellular clustering and nuclear molding were seen only in small cell carcinoma, whereas prominent nucleoli and an inflammatory background or diathesis were noted in lymphoma and melanoma cases. Intracellular pigment and multinucleation were recorded in melanoma cases.

CONCLUSION

The presence of small atypical cells in liquid-based UTCy should raise the suspicion of underlying malignancy involving the urinary tract. Cell clustering, nuclear molding, and hyperchromasia are helpful hints for the diagnosis of small cell carcinoma and the presence of small atypical cells with prominent nucleoli raises the possibility of lymphoma or melanoma.

Cytomorphological characteristics of low-grade papillary urothelial carcinoma for differential diagnosis from benign papillary urothelial lesions: logistic regression analysis in SurePath(™) liquid-based voided urine cytology

OBJECTIVE

The diagnosis of low-grade papillary urothelial carcinoma (LGPUC) in urine cytology specimens is challenging because of its subtle, minimally atypical findings. Furthermore, as SurePath(™) liquid-based cytology (LBC) is becoming a widely used method in urine cytology, the inevitable cytomorphological alterations resulting from this technique call for new morphological diagnostic criteria in LGPUC.

METHODS

Logistic regression analysis was carried out on SurePath slides from surgically proven voided urine specimens. The study was designed to include a test set (n = 141) and a validation set (n = 61), and evaluated significant discriminative parameters between LGPUC and benign papillary urothelial neoplasm (BPUN).

RESULTS

Of the seven cytological findings that were found to have statistical significance in univariate analysis, five were found to be independent variables: loss of polarity of papillaroid clusters, irregular contours, absence of columnar cells, hobnail features and hyperchromasia. These independent variables had an area under the curve (AUC) of 0.781.

CONCLUSIONS

The distinctive cytological criteria identified above may prove to be helpful in cases in which other conventional criteria for LGPUC are insufficient for diagnosis.

Myeloid leukemia in a urine specimen: a case report and review of the literature

Urinary tract cytology has a long history of utilization for the diagnosis and follow-up of tumors involving the urothelial tract. As expected, the most common tumor encountered in exfoliative urine cytology is urothelial carcinoma. While the sensitivity of urinary tract cytology for the diagnosis of low-grade urothelial carcinomas is low, its sensitivity and accuracy for high grade urothelial carcinomas is much higher. However, nonurothelial malignancies, such as hematopoietic malignancies, can also be encountered in urine specimens. Leukemic cells in urine can be diagnosed readily by cytological examination in cases where more invasive procedures are difficult to perform. Additionally, cell block sections can be utilized to determine the immunocytochemical profile of the tumor cells to confirm the diagnosis. Herein we report a case of a 75-year-old man with a past medical history of acute myeloid leukemia (AML), who presented with congested heart failure and painless macroscopic hematuria. AML relapse was diagnosed. Cytological examination of the urine using a ThinPrep® smear, cytospin preparation, and immunohistochemical stains performed on the cell block sections were examined. Findings were consistent with leukemic cells of myeloid origin in the bladder washing specimen.

Imaging systems and urine specimens: a new match made?

The application of standard cytologic examination of the urine specimen can now be augmented by using ThinPrep technology and other automatic screening devices. The technical and preparation adjustments required for these new technologies notwithstanding, the potential for shorter screening times and the possibility of greater diagnostic accuracy may provide great diagnostic promise for this very common specimen.

Selective screening for nongynecologic cytology specimens: modifying the screening process for improved efficiency and practice

Nongynecologic (NG) cytology cases usually generate multiple slides. In cases showing overtly malignant or neoplastic cells, the cytotechnologist (CT) may not need to screen all slides. We test the hypothesis that selective screening of a subset of slides by the CT is as effective as "routine screening" of all slides. The selective screening process (SSP) was performed by having the cytotechnologist (CT) screen and mark overtly malignant or neoplastic cells on up to three slides. Cases requiring more slides to be screened were not included in SSP. For each SSP case, the total slide count, number of slides screened, final diagnosis, and cytologic-histologic correlation (CHC) data were collected over 10 months and compared to the data from routinely screened cases. SSP was performed on 191 cases during a 10-month period. An average of 1.9 slides per case was screened by the cytotechnologist using the SSP. The average number of unscreened slides passed to the pathologist was 6.3 per case. On average, SSP resulted in 83.6 min of CT's time saved per day. Quality control by CHC demonstrated no false-positive cases in either the SSP or "routinely screened" groups. The diagnostic accuracy of the specific cytology diagnoses was 98% by SSP and 100% by "routine screening." SSP provides a mechanism for the cytotechnologist to "screen" fewer slides and pass the cases to the pathologist more efficiently without compromising overall patient care.

Comparison of ThinPrep preparations to other preparation types in gastrointestinal cytology: observations from the College of American Pathologists Interlaboratory Comparison Program in Nongynecologic Cytology

CONTEXT

Differences in participant responses for ThinPrep (TP) and non-ThinPrep (NTP) preparations for gastrointestinal cytology challenges, which circulated in the College of American Pathologists Interlaboratory Comparison Program in Nongynecologic Cytology (CAP NGC), may help to identify performance variations between preparation types.

OBJECTIVE

To compare the performance of TP-prepared slides of gastrointestinal exfoliative cytology specimens to that of NTP preparations in the CAP NGC program.

DESIGN

Participant responses between 2000 and 2007 were evaluated for esophageal wash/brush, gastric wash/brush, and biliary tract brush specimens with a reference diagnosis of adenocarcinoma, squamous cell carcinoma, carcinoid, or spindle cell neoplasm. ThinPrep challenges were compared with NTP preparations (conventional smears, cytospins) for discordant responses.

RESULTS

In all, 6023 pathologist responses and 3825 cytotechnologist responses were reviewed. Non-ThinPrep preparations comprised 93% (n = 11 588) of the challenges, while 7% (n = 912) were TP material. A match for a "positive/suspicious" diagnosis was seen in 88.5% of NTP and 95.9% of TP preparations (P < .001). These results were statistically significant when the specific reference diagnosis was adenocarcinoma (P < .001). Overall performance of cytotechnologists was not different from that of pathologists (89.2% versus 89.0%; P = .75). Cytotechnologists had better performance for detecting squamous cell carcinoma (96.3% versus 92.6%; P < .001), while pathologists had better performance for detecting spindle cell neoplasm (79.7% versus 42.9%; P < .001).

CONCLUSIONS

ThinPrep preparations performed significantly better than NTP preparations in gastrointestinal cytology specimens circulated in an interlaboratory comparison program. Performance varied by reference interpretation, with the best performance for the interpretation of adenocarcinoma. Cytotechnologists and pathologists performed at the same level overall, but with differences for the diagnosis of spindle cell neoplasm and squamous carcinoma.