Deep learning based digital cell profiles for risk stratification of urine cytology images

Comparative evaluation of slide scanners, scan settings, and cytopreparations for digital urine cytology

Background

Acquiring well-focused digital images of cytology slides with scanners can be challenging due to the 3-dimensional nature of the slides. This study evaluates performances of whole-slide images (WSIs) obtained from 2 different cytopreparations by 2 distinct scanners with 3 focus modes.

Methods

Fourteen urine specimens were collected from patients with urothelial carcinoma. Each specimen was equally divided into 2 portions, prepared with Cytospin and ThinPrep methods and scanned for WSIs using Leica (Aperio AT2) and Hamamatsu (NanoZoomer S360) scanners, respectively. The scan settings included 3 focus modes (default, semi-auto, and manual) for single-layer scanning, along with a manual focus mode for 21 Z-layers scanning. Performance metrics were evaluated including scanning success rate, artificial intelligence (AI) algorithm-inferred atypical cell numbers and coverage rate (atypical cell numbers in single or multiple Z-layers divided by the total atypical cell numbers in 21 Z-layers), scanning time, and image file size.

Results

The default mode had scanning success rates of 85.7% or 92.9%, depending on the scanner used. The semi-auto mode increased success to 92.9% or 100%, and manual even further to 100%. However, these changes did not affect the standardized median atypical cell numbers and coverage rates. The selection of scanners, cytopreparations, and Z-stacking influenced standardized median atypical cell numbers and coverage rates, scanning times, and image file sizes.

Discussion

Both scanners showed satisfactory scanning. We recommend using semi-auto or manual focus modes to achieve a scanning success rate of up to 100%. Additionally, a minimum of 9-layer Z-stacking at 1 μm intervals is required to cover 80% of atypical cells. These advanced focus methods do not impact the number of atypical cells or their coverage rate. While Z-stacking enhances the AI algorithm's inferred quantity and coverage rates of atypical cells, it simultaneously results in longer scanning times and larger image file sizes.

Computer-assisted urine cytology: Faster, cheaper, better?

Recent advancements in computer-assisted diagnosis (CAD) have catalysed significant progress in pathology, particularly in the realm of urine cytopathology. This review synthesizes the latest developments and challenges in CAD for diagnosing urothelial carcinomas, addressing the limitations of traditional urinary cytology. Through a literature review, we identify and analyse CAD models and algorithms developed for urine cytopathology, highlighting their methodologies and performance metrics. We discuss the potential of CAD to improve diagnostic accuracy, efficiency and patient outcomes, emphasizing its role in streamlining workflow and reducing errors. Furthermore, CAD tools have shown potential in exploring pathological conditions, uncovering novel biomarkers and prognostic/predictive features previously unknown or unseen. Finally, we examine the practical issues surrounding the integration of CAD into clinical practice, including regulatory approval, validation and training for pathologists. Despite the promising results, challenges remain, necessitating further research and validation efforts. Overall, CAD presents a transformative opportunity to revolutionize diagnostic practices in urine cytopathology, paving the way for enhanced patient care and outcomes.

CytoGAN: Unpaired staining transfer by structure preservation for cytopathology image analysis

With the development of digital pathology, deep learning is increasingly being applied to endometrial cell morphology analysis for cancer screening. And cytology images with different staining may degrade the performance of these analysis algorithms. To address the impact of staining patterns, many strategies have been proposed and hematoxylin and eosin (H&E) images have been transferred to other staining styles. However, none of the existing methods are able to generate realistic cytological images with preserved cellular layout, and many important clinical structural information is lost. To address the above issues, we propose a different staining transformation model, CytoGAN, which can quickly and realistically generate images with different staining styles. It includes a novel structure preservation module that preserves the cell structure well, even if the resolution or cell size between the source and target domains do not match. Meanwhile, a stain adaptive module is designed to help the model generate realistic and high-quality endometrial cytology images. We compared our model with ten state-of-the-art stain transformation models and evaluated by two pathologists. Furthermore, in the downstream endometrial cancer classification task, our algorithm improves the robustness of the classification model on multimodal datasets, with more than 20 % improvement in accuracy. We found that generating specified specific stains from existing H&E images improves the diagnosis of endometrial cancer. Our code will be available on github.

Segmentation, feature extraction and classification of leukocytes leveraging neural networks, a comparative study

The gold standard of leukocyte differentiation is a manual examination of blood smears, which is not only time and labor intensive but also susceptible to human error. As to automatic classification, there is still no comparative study of cell segmentation, feature extraction, and cell classification, where a variety of machine and deep learning models are compared with home-developed approaches. In this study, both traditional machine learning of K-means clustering versus deep learning of U-Net, U-Net + ResNet18, and U-Net + ResNet34 were used for cell segmentation, producing segmentation accuracies of 94.36% versus 99.17% for the dataset of CellaVision and 93.20% versus 98.75% for the dataset of BCCD, confirming that deep learning produces higher performance than traditional machine learning in leukocyte classification. In addition, a series of deep-learning approaches, including AlexNet, VGG16, and ResNet18, was adopted to conduct feature extraction and cell classification of leukocytes, producing classification accuracies of 91.31%, 97.83%, and 100% of CellaVision as well as 81.18%, 91.64% and 97.82% of BCCD, confirming the capability of the increased deepness of neural networks in leukocyte classification. As to the demonstrations, this study further conducted cell-type classification of ALL-IDB2 and PCB-HBC datasets, producing high accuracies of 100% and 98.49% among all literature, validating the deep learning model used in this study.

LESS: Label-efficient multi-scale learning for cytological whole slide image screening

In computational pathology, multiple instance learning (MIL) is widely used to circumvent the computational impasse in giga-pixel whole slide image (WSI) analysis. It usually consists of two stages: patch-level feature extraction and slide-level aggregation. Recently, pretrained models or self-supervised learning have been used to extract patch features, but they suffer from low effectiveness or inefficiency due to overlooking the task-specific supervision provided by slide labels. Here we propose a weakly-supervised Label-Efficient WSI Screening method, dubbed LESS, for cytological WSI analysis with only slide-level labels, which can be effectively applied to small datasets. First, we suggest using variational positive-unlabeled (VPU) learning to uncover hidden labels of both benign and malignant patches. We provide appropriate supervision by using slide-level labels to improve the learning of patch-level features. Next, we take into account the sparse and random arrangement of cells in cytological WSIs. To address this, we propose a strategy to crop patches at multiple scales and utilize a cross-attention vision transformer (CrossViT) to combine information from different scales for WSI classification. The combination of our two steps achieves task-alignment, improving effectiveness and efficiency. We validate the proposed label-efficient method on a urine cytology WSI dataset encompassing 130 samples (13,000 patches) and a breast cytology dataset FNAC 2019 with 212 samples (21,200 patches). The experiment shows that the proposed LESS reaches 84.79%, 85.43%, 91.79% and 78.30% on the urine cytology WSI dataset, and 96.88%, 96.86%, 98.95%, 97.06% on the breast cytology high-resolution-image dataset in terms of accuracy, AUC, sensitivity and specificity. It outperforms state-of-the-art MIL methods on pathology WSIs and realizes automatic cytological WSI cancer screening.

A Fully Automated Artificial Intelligence System to Assist Pathologists' Diagnosis to Predict Histologically High-grade Urothelial Carcinoma from Digitized Urine Cytology Slides Using Deep Learning

BACKGROUND

Urine cytology, although a useful screening method for urothelial carcinoma, lacks sensitivity. As an emerging technology, artificial intelligence (AI) improved image analysis accuracy significantly.

OBJECTIVE

To develop a fully automated AI system to assist pathologists in the histological prediction of high-grade urothelial carcinoma (HGUC) from digitized urine cytology slides.

DESIGN, SETTING, AND PARTICIPANTS

We digitized 535 consecutive urine cytology slides for AI use. Among these slides, 181 were used for AI development, 39 were used as AI test data to identify HGUC by cell-level classification, and 315 were used as AI test data for slide-level classification.

OUTCOME MEASUREMENTS AND STATISTICAL ANALYSIS

Out of the 315 slides, 171 were collected immediately prior to bladder biopsy or transurethral resection of bladder tumor, and then outcomes were compared with the histological presence of HGUC in the surgical specimen. The primary aim was to compare AI prediction of the histological presence of HGUC with the pathologist's histological diagnosis of HGUC. Secondary aims were to compare the time required for AI evaluation and concordance between the AI's classification and pathologist's cytology diagnosis.

RESULTS AND LIMITATIONS

The AI capability for predicting the histological presence of HGUC was 0.78 for the area under the curve. Comparing the AI predictive performance with pathologists' diagnosis, the AI sensitivity of 63% for histological HGUC prediction was superior to a pathologists' cytology sensitivity of 46% (p = 0.0037). On the contrary, there was no significant difference between the AI specificity of 83% and pathologists' specificity of 89% (p = 0.13), and AI accuracy of 74% and pathologists' accuracy of 68% (p = 0.08). The time required for AI evaluation was 139 s. With respect to the concordance between the AI prediction and pathologist's cytology diagnosis, the accuracy was 86%. Agreements with positive and negative findings were 92% and 84%, respectively.

CONCLUSIONS

We developed a fully automated AI system to assist pathologists' histological diagnosis of HGUC using digitized slides. This AI system showed significantly higher sensitivity than a board-certified cytopathologist and may assist pathologists in making urine cytology diagnoses, reducing their workload.

PATIENT SUMMARY

In this study, we present a deep learning-based artificial intelligence (AI) system that classifies urine cytology slides according to the Paris system. An automated AI system was developed and validated with 535 consecutive urine cytology slides. The AI predicted histological high-grade urothelial carcinoma from digitized urine cytology slides with superior sensitivity than pathologists, while maintaining comparable specificity and accuracy.

Artificial-Intelligence-Assisted Detection of Metastatic Colorectal Cancer Cells in Ascitic Fluid

Ascites cytology is a cost-effective test for metastatic colorectal cancer (CRC) in the abdominal cavity. However, metastatic carcinoma of the peritoneum is difficult to diagnose based on biopsy findings, and ascitic aspiration cytology has a low sensitivity and specificity and a high inter-observer variability. The aim of the present study was to apply artificial intelligence (AI) to classify benign and malignant cells in ascites cytology patch images of metastatic CRC using a deep convolutional neural network. Datasets were collected from The OPEN AI Dataset Project, a nationwide cytology dataset for AI research. The numbers of patch images used for training, validation, and testing were 56,560, 7068, and 6534, respectively. We evaluated 1041 patch images of benign and metastatic CRC in the ascitic fluid to compare the performance of pathologists and an AI algorithm, and to examine whether the diagnostic accuracy of pathologists improved with the assistance of AI. This AI method showed an accuracy, a sensitivity, and a specificity of 93.74%, 87.76%, and 99.75%, respectively, for the differential diagnosis of malignant and benign ascites. The diagnostic accuracy and sensitivity of the pathologist with the assistance of the proposed AI method increased from 86.8% to 90.5% and from 73.3% to 79.3%, respectively. The proposed deep learning method may assist pathologists with different levels of experience in diagnosing metastatic CRC cells of ascites.

Digital cytology part 2: artificial intelligence in cytology: a concept paper with review and recommendations from the American Society of Cytopathology Digital Cytology Task Force

Digital cytology and artificial intelligence (AI) are gaining greater adoption in the cytology laboratory. However, peer-reviewed real-world data and literature are lacking in regard to the current clinical landscape. The American Society of Cytopathology in conjunction with the International Academy of Cytology and the Digital Pathology Association established a special task force comprising 20 members with expertise and/or interest in digital cytology. The aim of the group was to investigate the feasibility of incorporating digital cytology, specifically cytology whole slide scanning and AI applications, into the workflow of the laboratory. In turn, the impact on cytopathologists, cytologists (cytotechnologists), and cytology departments were also assessed. The task force reviewed existing literature on digital cytology, conducted a worldwide survey, and held a virtual roundtable discussion on digital cytology and AI with multiple industry corporate representatives. This white paper, presented in 2 parts, summarizes the current state of digital cytology and AI practice in global cytology practice. Part 1 of the white paper is presented as a separate paper which details a review and best practice recommendations for incorporating digital cytology into practice. Part 2 of the white paper presented here provides a comprehensive review of AI in cytology practice along with best practice recommendations and legal considerations. Additionally, the cytology global survey results highlighting current AI practices by various laboratories, as well as current attitudes, are reported.

Examining longitudinal markers of bladder cancer recurrence through a semiautonomous machine learning system for quantifying specimen atypia from urine cytology

BACKGROUND

Urine cytology is generally considered the primary approach for screening for recurrence of bladder cancer. However, it is currently unclear how best to use cytological examinations for assessment and early detection of recurrence, beyond identifying a positive finding that requires more invasive methods to confirm recurrence and decide on therapeutic options. Because screening programs are frequent, and can be burdensome, finding quantitative means to reduce this burden for patients, cytopathologists, and urologists is an important endeavor and can improve both the efficiency and reliability of findings. Additionally, identifying ways to risk-stratify patients is crucial for improving quality of life while reducing the risk of future recurrence or progression of the cancer.

METHODS

In this study, a computational machine learning tool, AutoParis-X, was leveraged to extract imaging features from urine cytology examinations longitudinally to study the predictive potential of urine cytology for assessing recurrence risk. This study examined how the significance of imaging predictors changes over time before and after surgery to determine which predictors and time periods are most relevant for assessing recurrence risk.

RESULTS

Results indicate that imaging predictors extracted using AutoParis-X can predict recurrence as well or better than traditional cytological/histological assessments alone and that the predictiveness of these features is variable across time, with key differences in overall specimen atypia identified immediately before tumor recurrence.

CONCLUSIONS

Further research will clarify how computational methods can be effectively used in high-volume screening programs to improve recurrence detection and complement traditional modes of assessment.

Artificial Intelligence in the Advanced Diagnosis of Bladder Cancer-Comprehensive Literature Review and Future Advancement

Artificial intelligence is highly regarded as the most promising future technology that will have a great impact on healthcare across all specialties. Its subsets, machine learning, deep learning, and artificial neural networks, are able to automatically learn from massive amounts of data and can improve the prediction algorithms to enhance their performance. This area is still under development, but the latest evidence shows great potential in the diagnosis, prognosis, and treatment of urological diseases, including bladder cancer, which are currently using old prediction tools and historical nomograms. This review focuses on highly significant and comprehensive literature evidence of artificial intelligence in the management of bladder cancer and investigates the near introduction in clinical practice.

Deep learning for computational cytology: A survey

Computational cytology is a critical, rapid-developing, yet challenging topic in medical image computing concerned with analyzing digitized cytology images by computer-aided technologies for cancer screening. Recently, an increasing number of deep learning (DL) approaches have made significant achievements in medical image analysis, leading to boosting publications of cytological studies. In this article, we survey more than 120 publications of DL-based cytology image analysis to investigate the advanced methods and comprehensive applications. We first introduce various deep learning schemes, including fully supervised, weakly supervised, unsupervised, and transfer learning. Then, we systematically summarize public datasets, evaluation metrics, versatile cytology image analysis applications including cell classification, slide-level cancer screening, nuclei or cell detection and segmentation. Finally, we discuss current challenges and potential research directions of computational cytology.

Uncovering additional predictors of urothelial carcinoma from voided urothelial cell clusters through a deep learning-based image preprocessing technique

BACKGROUND

Urine cytology is commonly used as a screening test for high-grade urothelial carcinoma for patients with risk factors or hematuria and is an essential step in longitudinal monitoring of patients with previous bladder cancer history. However, the semisubjective nature of current reporting systems for urine cytology (e.g., The Paris System) can hamper reproducibility. For instance, the incorporation of urothelial cell clusters into the classification schema is still an item of debate and perplexity among expert cytopathologists because several previous works have disputed their diagnostic relevance.

METHODS

In this work, an automated preprocessing tool for urothelial cell cluster assessment was developed that divides urothelial cell clusters into meaningful components for downstream assessment (ie, population-based studies, workflow automation).

RESULTS

In this work, an automated preprocessing tool for urothelial cell cluster assessment was developed that divides urothelial cell clusters into meaningful components for downstream assessment (ie, population-based studies, workflow automation). Results indicate that cell cluster atypia (i.e., defined by whether the cell cluster harbored multiple atypical cells, thresholded by a minimum number of cells), cell border overlap and smoothness, and total number of clusters are important markers of specimen atypia when considering assessment of urothelial cell clusters.

CONCLUSIONS

Markers established through techniques to separate cell clusters may have wider applicability for the design and implementation of machine learning approaches for urine cytology assessment.

Recent Application of Artificial Intelligence in Non-Gynecological Cancer Cytopathology: A Systematic Review

Simple Summary

Artificial intelligence (AI) has attracted significant interest in the healthcare sector due to its promising results. Cytological examination is a critical step in the initial diagnosis of cancer. Here, we conducted a systematic review with quantitative analysis to understand the current status of AI applications in non-gynecological (non-GYN) cancer cytology. In our analysis, we found that most of the studies focused on classification and segmentation tasks. Overall, AI showed promising results for non-GYN cancer cytopathology analysis. However, the lack of well-annotated, large-scale datasets with Z-stacking and external cross-validation was the major limitation across all studies.

Abstract

State-of-the-art artificial intelligence (AI) has recently gained considerable interest in the healthcare sector and has provided solutions to problems through automated diagnosis. Cytological examination is a crucial step in the initial diagnosis of cancer, although it shows limited diagnostic efficacy. Recently, AI applications in the processing of cytopathological images have shown promising results despite the elementary level of the technology. Here, we performed a systematic review with a quantitative analysis of recent AI applications in non-gynecological (non-GYN) cancer cytology to understand the current technical status. We searched the major online databases, including MEDLINE, Cochrane Library, and EMBASE, for relevant English articles published from January 2010 to January 2021. The searched query terms were: “artificial intelligence”, “image processing”, “deep learning”, “cytopathology”, and “fine-needle aspiration cytology.” Out of 17,000 studies, only 26 studies (26 models) were included in the full-text review, whereas 13 studies were included for quantitative analysis. There were eight classes of AI models treated of according to target organs: thyroid (n = 11, 39%), urinary bladder (n = 6, 21%), lung (n = 4, 14%), breast (n = 2, 7%), pleural effusion (n = 2, 7%), ovary (n = 1, 4%), pancreas (n = 1, 4%), and prostate (n = 1, 4). Most of the studies focused on classification and segmentation tasks. Although most of the studies showed impressive results, the sizes of the training and validation datasets were limited. Overall, AI is also promising for non-GYN cancer cytopathology analysis, such as pathology or gynecological cytology. However, the lack of well-annotated, large-scale datasets with Z-stacking and external cross-validation was the major limitation found across all studies. Future studies with larger datasets with high-quality annotations and external validation are required.

Evaluation of an artificial intelligence algorithm for assisting the Paris System in reporting urinary cytology: A pilot study

BACKGROUND

The Paris System for Reporting Urinary Cytology (TPS) has been shown to improve bladder cancer diagnosis. Advances in artificial intelligence (AI) may assist and improve the clinical workflow by applying TPS in routine diagnostic services.

METHODS

A deep-learning-based algorithm was developed to identify urothelial cancer candidate cells using whole-slide images (WSIs). In the testing cohort, 131 urine cytology slides were retrospectively retrieved and analyzed using this AI algorithm. The authors compared the performance of one cytopathologist and two cytotechnologists using AI-assisted digital urine cytology. Then, the AI-assisted WSIs were evaluated in the clinical workflow. The cytopathologist first made a diagnosis by reviewing the AI-inferred WSIs and quantitative data (nuclear-to-cytoplasmic ratio and nuclear size) for each sample. After a washout period, the same cytopathologist made a diagnosis for the same samples using direct microscopy. All diagnosis results were compared with the expert panel consensus.

RESULTS

The AI-assisted diagnosis by the two cytotechnologists and the one cytopathologist demonstrated performance results that were comparable to the expert panel consensus (sensitivity, 79.5% and 82.1% vs. 92.3%, respectively; specificity, 100% and 98.9% vs. 100%, respectively). Furthermore, the performance of the AI-assisted WSIs compared with the microscopic diagnosis by the cytopathologist demonstrated superior sensitivity (92.3% vs. 87.2%) and negative predictive value (96.8% vs. 94.8%). In addition, the AI-assisted reporting demonstrated near perfect agreement with the expert panel consensus (κ = 0.944) and the microscopic diagnosis (κ = 0.862).

CONCLUSIONS

The AI algorithm developed by the authors effectively assisted TPS-based reporting by providing AI-inferred WSIs and quantitative data.

Urine cell image recognition using a deep-learning model for an automated slide evaluation system

OBJECTIVES

To develop a classification system for urine cytology with artificial intelligence (AI) using a convolutional neural network algorithm that classifies urine cell images as negative (benign) or positive (atypical or malignant).

PATIENTS AND METHODS

We collected 195 urine cytology slides from consecutive patients with a histologically confirmed diagnosis of urothelial cancer (between January 2016 and December 2017). Two certified cytotechnologists independently evaluated and labelled each slide; 4637 cell images with concordant diagnoses were selected, including 3128 benign cells (negative), 398 atypical cells, and 1111 cells that were malignant or suspicious for malignancy (positive). This pathologically confirmed labelled dataset was used to represent the ground truth for AI training/validation/testing. Customized CutMix (CircleCut) and Refined Data Augmentation were used for image processing. The model architecture included EfficientNet B6 and Arcface. We used 80% of the data for training and validation (4:1 ratio) and 20% for testing. Model performance was evaluated with fivefold cross-validation. A receiver-operating characteristic (ROC) analysis was used to evaluate the binary classification model. Bayesian posterior probabilities for the AI performance measure (Y) and cytotechnologist performance measure (X) were compared.

RESULTS

The area under the ROC curve was 0.99 (95% confidence interval [CI] 0.98-0.99), the highest accuracy was 95% (95% CI 94-97), sensitivity was 97% (95% CI 95-99), and specificity was 95% (95% CI 93-97). The accuracy of AI surpassed the highest level of cytotechnologists for the binary classification [Pr(Y > X) = 0.95]. AI achieved >90% accuracy for all cell subtypes. In the subgroup analysis based on the clinicopathological characteristics of patients who provided the test cells, the accuracy of AI ranged between 89% and 97%.

CONCLUSION

Our novel AI classification system for urine cytology successfully classified all cell subtypes with an accuracy of higher than 90%, and achieved diagnostic accuracy of malignancy superior to the highest level achieved by cytotechnologists.