Siamese deep learning video flow cytometry for automatic and label-free clinical cervical cancer cell analysis.

Automatic and label-free screening methods may help to reduce cervical cancer mortality rates, especially in developing regions. The latest advances of deep learning in the biomedical optics field provide a more automatic approach to solving clinical dilemmas. However, existing deep learning methods face challenges, such as the requirement of manually annotated training sets for clinical sample analysis. Here, we develop Siamese deep learning video flow cytometry for the analysis of clinical cervical cancer cell samples in a smear-free manner. High-content light scattering images of label-free single cells are obtained via the video flow cytometer. Siamese deep learning, a self-supervised method, is built to introduce cell lineage cells into an analysis of clinical cells, which utilizes generated similarity metrics as label annotations for clinical cells. Compared with other deep learning methods, Siamese deep learning achieves a higher accuracy of up to 87.11%, with about 5.62% improvement for label-free clinical cervical cancer cell classification. The Siamese deep learning video flow cytometry demonstrated here is promising for automatic, label-free analysis of many types of cells from clinical samples without cell smears.

Label-Free Light Scattering Imaging with Purified Brownian Motion Differentiates Small Extracellular Vesicles in Cell Microenvironments.

Small extracellular vesicles (sEVs) are heterogeneous biological nanoparticles (NPs) with wide biomedicine applications. Tracking individual nanoscale sEVs can reveal information that conventional microscopic methods may lack, especially in cellular microenvironments. This usually requires biolabeling to identify single sEVs. Here, we developed a light scattering imaging method based on dark-field technology for label-free nanoparticle diffusion analysis (NDA). Compared with nanoparticle tracking analysis (NTA), our method was shown to determine the diffusion probabilities of a single NP. It was demonstrated that accurate size determination of NPs of 41 and 120 nm in diameter is achieved by purified Brownian motion (pBM), without or within the cell microenvironments. Our pBM method was also shown to obtain a consistent size estimation of the normal and cancerous plasma-derived sEVs without and within cell microenvironments, while cancerous plasma-derived sEVs are statistically smaller than normal ones. Moreover, we showed that the velocity and diffusion coefficient are key parameters for determining the diffusion types of the NPs and sEVs in a cancerous cell microenvironment. Our light scattering-based NDA and pBM methods can be used for size determination of NPs, even in cell microenvironments, and also provide a tool that may be used to analyze sEVs for many biomedical applications.

Three-dimensional deep regression-based light scattering imaging system for nanoscale exosome analysis.

Exosomes are extracellular vesicles that serve as promising intrinsic nanoscale biomarkers for disease diagnosis and treatment. Nanoparticle analysis technology is widely used in the field of exosome study. However, the common particle analysis methods are usually complex, subjective, and not robust. Here, we develop a three-dimensional (3D) deep regression-based light scattering imaging system for nanoscale particle analysis. Our system solves the problem of object focusing in common methods and acquires light scattering images of label-free nanoparticles as small as 41 nm in diameter. We develop a new method for nanoparticle sizing with 3D deep regression, where the 3D time series Brownian motion data of single nanoparticles are input as a whole, and sizes are output automatically for both entangled and untangled nanoparticles. Exosomes from the normal and cancer liver cell lineage cells are observed and automatically differentiated by our system. The 3D deep regression-based light scattering imaging system is expected to be widely used in the field of nanoparticle analysis and nanomedicine.

High-content video flow cytometry with digital cell filtering for label-free cell classification by machine learning.

Recent development of imaging flow cytometry (IFC) has enabled the measurements of single cells with high throughput, where fluorescent labels provide specificity for cellular diagnosis. The fluorescent labels may disturb the cell functions, and the requirements for high-throughput measurements limit the cell image quality. Here, we develop the high-content video flow cytometry (VFC) that measures unlabeled single cells with a rate of approximately 1000 cells per minute. For the obtained big data, the frame of interest (FOI) is automatically prepared by a digital cell filtering technique with machine learning. Cervical carcinoma cell lines (Caski, HeLa and C33-A cells) are differentiated with an accuracy of 91.5%, 90.5%, and 90.5% by deep learning in a three-way classification, respectively. The high-content VFC not only provides high-quality images of single cells with high throughput and rewinding, but also performs automatic digital cell filtering and label-free cell classification that may have clinical applications.

Differentiating single cervical cells by mitochondrial fluorescence imaging and deep learning-based label-free light scattering with multi-modal static cytometry.

Cervical cancer is a high-risk disease that threatens women's health globally. In this study, we developed the multi-modal static cytometry that adopted different features to classify the typical human cervical epithelial cells (H8) and cervical cancer cells (HeLa). With the light-sheet static cytometry, we obtain brightfield (BF) images, fluorescence (FL) images and two-dimensional (2D) light scattering (LS) patterns of single cervical cells. Three feature extraction methods are used to extract multi-modal features based on different data characteristics. Analysis and classification of morphological and textural features demonstrate the potential of intracellular mitochondria in cervical cancer cell classification. The deep learning method is used to automatically extract deep features of label-free LS patterns, and an accuracy of 76.16% for the classification of the above two kinds of cervical cells is obtained, which is higher than the other two single modes (BF and FL). Our multi-modal static cytometry uses a variety of feature extraction and analysis methods to provide the mitochondria as promising internal biomarkers for cervical cancer diagnosis, and to show the promise of label-free, automatic classification of early cervical cancer with deep learning-based 2D light scattering.

A microfluidic cytometer for white blood cell analysis.

Despite the wide use of cytometry for white blood cell classification, the performance of traditional cytometers in point-of-care testing remains to be improved. Microfluidic techniques have been shown with considerable potentials in the development of portable devices. Here we present a prototype of microfluidic cytometer which integrates a three-dimensional hydrodynamic focusing system and an on-chip optical system to count and classify white blood cells. By adjusting the flow speed of sheath flow and sample flow, the blood cells can be horizontally and vertically focused in the center of microchannel. Optical fibers and on-chip microlens are embedded for the excitation and detection of single-cell. The microfluidic chip was validated by classifying white blood cells from clinical blood samples.

Risk Stratification of Early-Stage Cervical Cancer with Intermediate-Risk Factors: Model Development and Validation Based on Machine Learning Algorithm.

BACKGROUND: Adjuvant therapy for patients with cervical cancer (CC) with intermediate-risk factors remains controversial. The objectives of the present study are to assess the prognoses of patients with early-stage CC with pathological intermediate-risk factors and to provide a reference for adjuvant therapy choice. MATERIALS AND METHODS: This retrospective study included 481 patients with stage IB-IIA CC. Cox proportional hazards regression analysis, machine learning (ML) algorithms, Kaplan-Meier analysis, and the area under the receiver operating characteristic curve (AUC) were used to develop and validate prediction models for disease-free survival (DFS) and overall survival (OS). RESULTS: A total of 35 (7.3%) patients experienced recurrence, and 20 (4.2%) patients died. Two prediction models were built for DFS and OS using clinical information, including age, lymphovascular space invasion, stromal invasion, tumor size, and adjuvant treatment. Patients were divided into high-risk or low-risk groups according to the risk score cutoff value. The Kaplan-Meier analysis showed significant differences in DFS (p = .001) and OS (p = .011) between the two risk groups. In the traditional Sedlis criteria groups, there were no significant differences in DFS or OS (p > .05). In the ML-based validation, the best AUCs of DFS at 2 and 5 years were 0.69/0.69, and the best AUCs of OS at 2 and 5 years were 0.88/0.63. CONCLUSION: Two prognostic assessment models were successfully established, and risk grouping stratified the prognostic risk of patients with CC with pathological intermediate-risk factors. Evaluation of long-term survival will be needed to corroborate these findings. IMPLICATIONS FOR PRACTICE: The Sedlis criteria are intermediate-risk factors used to guide postoperative adjuvant treatment in patients with cervical cancer. However, for patients meeting the Sedlis criteria, the choice of adjuvant therapy remains controversial. This study developed two prognostic models based on pathological intermediate-risk factors. According to the risk score obtained by the prediction model, patients can be further divided into groups with high or low risk of recurrence and death. The prognostic models developed in this study can be used in clinical practice to stratify prognostic risk and provide more individualized adjuvant therapy choices to patients with early-stage cervical cancer.

Light scattering pattern specific convolutional network static cytometry for label-free classification of cervical cells.

Cervical cancer is a major gynecological malignant tumor that threatens women's health. Current cytological methods have certain limitations for cervical cancer early screening. Light scattering patterns can reflect small differences in the internal structure of cells. In this study, we develop a light scattering pattern specific convolutional network (LSPS-net) based on deep learning algorithm and integrate it into a 2D light scattering static cytometry for automatic, label-free analysis of single cervical cells. An accuracy rate of 95.46% for the classification of normal cervical cells and cancerous ones (mixed C-33A and CaSki cells) is obtained. When applied for the subtyping of label-free cervical cell lines, we obtain an accuracy rate of 93.31% with our LSPS-net cytometric technique. Furthermore, the three-way classification of the above different types of cells has an overall accuracy rate of 90.90%, and comparisons with other feature descriptors and classification algorithms show the superiority of deep learning for automatic feature extraction. The LSPS-net static cytometry may potentially be used for cervical cancer early screening, which is rapid, automatic and label-free.

Deep learning-based light scattering microfluidic cytometry for label-free acute lymphocytic leukemia classification.

The subtyping of Acute lymphocytic leukemia (ALL) is important for proper treatment strategies and prognosis. Conventional methods for manual blood and bone marrow testing are time-consuming and labor-intensive, while recent flow cytometric immunophenotyping has the limitations such as high cost. Here we develop the deep learning-based light scattering imaging flow cytometry for label-free classification of ALL. The single ALL cells confined in three dimensional (3D) hydrodynamically focused stream are excited by light sheet. Our label-free microfluidic cytometry obtains big-data two dimensional (2D) light scattering patterns from single ALL cells of B/T subtypes. A deep learning framework named Inception V3-SIFT (Scale invariant feature transform)-Scattering Net (ISSC-Net) is developed, which can perform high-precision classification of T-ALL and B-ALL cell line cells with an accuracy of 0.993 ± 0.003. Our deep learning-based 2D light scattering flow cytometry is promising for automatic and accurate subtyping of un-stained ALL.

Prognostic Assessment of Cervical Cancer Patients by Clinical Staging and Surgical-Pathological Factor: A Support Vector Machine-Based Approach.

Introduction: The International Federation of Gynecology and Obstetrics (FIGO) staging system is considered the most powerful prognostic factor in patients with cervical cancer. In addition, other surgical-pathological risk factors have been demonstrated to have significance in predicting the prognosis of patients. Therefore, the purpose of this study was to investigate the effects of the FIGO staging system and surgical-pathological risk factors on the prognosis of cervical cancer patients. Methods: A retrospective study was performed on patients diagnosed with cervical cancer at FIGO stage IB1-IIA2. Kaplan-Meier, Cox proportional hazards regression analysis and the support vector machine (SVM) algorithm were used to assess and validate the high-risk factors related to recurrence and death. Results: A total of 647 patients were included. Kaplan-Meier analysis showed that five high-risk factors, including FIGO stage, status of pelvic lymph node, parametrial involvement, tumor size, and depth of cervical cancer, had a significant effect on the prognosis of patients. In multivariate analysis, pelvic lymph node metastasis (hazard ratio [HR] 2.415, 95% confidence interval [CI] 1.471-3.965), parametrial involvement (HR 2.740, 95% CI 1.092-6.872) and >2/3 depth of cervical invasion (HR 2.263, 95% CI 1.045-4.902) were three independent risk factors of disease-free survival. Pelvic lymph node metastasis (HR 3.855, 95% CI 2.125-6.991) and parametrial involvement (HR 3.871, 95% CI 1.375-10.900) were two independent risk factors for overall survival. When all five high-risk factors were assembled and used for classification prediction through SVM, it achieved the highest prediction accuracy of recurrence (accuracy = 69.1%). The highest prediction accuracy for survival was 94.3% when only using the two independent predictors (the pathological status of lymph nodes and parametrium involvement) by SVM classifiers. Among the 13 groups of intermediate-risk factor, the combination of tumor size, histology and grade of differentiation was more accurate in predicting prognosis than the intermediate-risk factors in the Sedlis criteria (recurrence: 86.8% vs. 60.0%; death: 92.0% vs. 71.6%). Conclusions: The combination of FIGO stage and surgical-pathological risk factors can further enhance the prediction accuracy of the prognosis in patients with early-stage cervical cancer. Histology and grade of differentiation can further improve the prediction accuracy of intermediate-risk factors in the Sedlis criteria.

Predicting the Risk of Adverse Events in Pregnant Women With Congenital Heart Disease.

Background Women with congenital heart disease are considered at high risk for adverse events. Therefore, we aim to establish 2 prediction models for mothers and their offspring, which can predict the risk of adverse events occurred in pregnant women with congenital heart disease. Methods and Results A total of 318 pregnant women with congenital heart disease were included; 213 women were divided into the development cohort, and 105 women were divided into the validation cohort. Least absolute shrinkage and selection operator was used for predictor selection. After validation, multivariate logistic regression analysis was used to develop the model. Machine learning algorithms (support vector machine, random forest, AdaBoost, decision tree, k-nearest neighbor, naïve Bayes, and multilayer perceptron) were used to further verify the predictive ability of the model. Forty-one (12.9%) women experienced adverse maternal events, and 93 (29.2%) neonates experienced adverse neonatal events. Seven high-risk factors were discovered in the maternal model, including New York Heart Association class, Eisenmenger syndrome, pulmonary hypertension, left ventricular ejection fraction, sinus tachycardia, arterial blood oxygen saturation, and pregnancy duration. The machine learning-based algorithms showed that the maternal model had an accuracy of 0.76 to 0.86 (area under the receiver operating characteristic curve=0.74-0.87) in the development cohort, and 0.72 to 0.86 (area under the receiver operating characteristic curve=0.68-0.80) in the validation cohort. Three high-risk factors were discovered in the neonatal model, including Eisenmenger syndrome, preeclampsia, and arterial blood oxygen saturation. The machine learning-based algorithms showed that the neonatal model had an accuracy of 0.75 to 0.80 (area under the receiver operating characteristic curve=0.71-0.77) in the development cohort, and 0.72 to 0.79 (area under the receiver operating characteristic curve=0.69-0.76) in the validation cohort. Conclusions Two prenatal risk assessment models for both adverse maternal and neonatal events were established, which might assist clinicians in tailoring precise management and therapy in pregnant women with congenital heart disease.

Deep Learning-Based Single-Cell Optical Image Studies.

Optical imaging technology that has the advantages of high sensitivity and cost-effectiveness greatly promotes the progress of nondestructive single-cell studies. Complex cellular image analysis tasks such as three-dimensional reconstruction call for machine-learning technology in cell optical image research. With the rapid developments of high-throughput imaging flow cytometry, big data cell optical images are always obtained that may require machine learning for data analysis. In recent years, deep learning has been prevalent in the field of machine learning for large-scale image processing and analysis, which brings a new dawn for single-cell optical image studies with an explosive growth of data availability. Popular deep learning techniques offer new ideas for multimodal and multitask single-cell optical image research. This article provides an overview of the basic knowledge of deep learning and its applications in single-cell optical image studies. We explore the feasibility of applying deep learning techniques to single-cell optical image analysis, where popular techniques such as transfer learning, multimodal learning, multitask learning, and end-to-end learning have been reviewed. Image preprocessing and deep learning model training methods are then summarized. Applications based on deep learning techniques in the field of single-cell optical image studies are reviewed, which include image segmentation, super-resolution image reconstruction, cell tracking, cell counting, cross-modal image reconstruction, and design and control of cell imaging systems. In addition, deep learning in popular single-cell optical imaging techniques such as label-free cell optical imaging, high-content screening, and high-throughput optical imaging cytometry are also mentioned. Finally, the perspectives of deep learning technology for single-cell optical image analysis are discussed. © 2020 International Society for Advancement of Cytometry.

Two-Dimensional Light Scattering Anisotropy Cytometry for Label-Free Classification of Ovarian Cancer Cells via Machine Learning.

We develop a single-mode fiber-based cytometer for the obtaining of two-dimensional (2D) light scattering patterns from static single cells. Anisotropy of the 2D light scattering patterns of single cells from ovarian cancer and normal cell lines is investigated by histograms of oriented gradients (HOG) method. By analyzing the HOG descriptors with support vector machine, an accuracy rate of 92.84% is achieved for the automatic classification of these two kinds of label-free cells. The 2D light scattering anisotropy cytometry combined with machine learning may provide a label-free, automatic method for screening of ovarian cancer cells, and other types of cells. © 2019 International Society for Advancement of Cytometry.

Automatic Classification of Label-Free Cells from Small Cell Lung Cancer and Poorly Differentiated Lung Adenocarcinoma with 2D Light Scattering Static Cytometry and Machine Learning.

Small cell lung cancer (SCLC) needs to be classified from poorly differentiated lung adenocarcinoma (PDLAC) for appropriate treatment of lung cancer patients. Currently, the classification is achieved by experienced clinicians, radiologists and pathologists based on subjective and qualitative analysis of imaging, cytological and immunohistochemical (IHC) features. Label-free classification of lung cancer cell lines is developed here by using two-dimensional (2D) light scattering static cytometric technique. Measurements of scattered light at forward scattering (FSC) and side scattering (SSC) by using conventional cytometry show that SCLC cells are overlapped with PDLAC cells. However, our 2D light scattering static cytometer reveals remarkable differences between the 2D light scattering patterns of SCLC cell lines (H209 and H69) and PDLAC cell line (SK-LU-1). By adopting support vector machine (SVM) classifier with leave-one-out cross-validation (LOO-CV), SCLC and PDLAC cells are automatically classified with an accuracy of 99.87%. Our label-free 2D light scattering static cytometer may serve as a new, accurate, and easy-to-use method for the automatic classification of SCLC and PDLAC cells. © 2018 International Society for Advancement of Cytometry.

Label-free light-sheet microfluidic cytometry for the automatic identification of senescent cells.

Label-free microfluidic cytometry is of increasing interest for single cell analysis due to its advantages of high-throughput, miniaturization, as well as noninvasive detection. Here we develop a next generation label-free light-sheet microfluidic cytometer for single cell analysis by two-dimensional (2D) light scattering measurements. Our cytometer integrates light sheet illumination with a disposable hydrodynamic focusing unit, which can achieve 3D hydrodynamic focusing of a sample fluid to a diameter of 19 micrometer without microfabrication. This integration also improves the signal to noise ratio (SNR) for the acquisition of 2D light scattering patterns from label-free cells. Particle sizing with submicron resolution is achieved by our light-sheet flow cytometer, where Euclidean distance-based similarity measures are performed. Label-free, automatic classification of senescent and normal cells is achieved with a high accuracy rate by incorporating our light-sheet flow cytometry with support vector machine (SVM) algorithms. Our light-sheet microfluidic cytometry with a microfabrication-free hydrodynamic focusing unit may find wide applications for automatic and label-free clinical diagnosis.

Hypocrellin B-loaded, folate-conjugated polymeric micelle for intraperitoneal targeting of ovarian cancer in vitro and in vivo.

Photodynamic therapy (PDT) is considered an innovative and attractive modality to treat ovarian cancer. In the present study, a biodegradable polymer poly (ethylene glycol) (PEG)-poly (lactic acid)(PLA)-folate (FA-PEG-PLA) was prepared in order to synthesize an active-targeting, water-soluble and pharmacomodulated photosensitizer nanocarrier. Drug-loading content, encapsulation efficiency, in vitro and in vivo release were characterized, in which hypocrellin B (HB)/FA-PEG-PLA micelles had a high encapsulation efficiency and much slower control release for drugs compared to free drugs (P < .05). To evaluate the targeting ability of the HB/FA-PEG-PLA micelles, a cellular uptake study in vitro was carried out, which showed significantly enhanced uptake of HB/FA-PEG-PLA micelles in SKOV3 (FR+) compared to A2780 cancer cells (FR-). The enhanced uptake of HB/FA-PEG-PLA micelles to cancer cells resulted in a more effective post-PDT killing of SKOV3 cells compared to plain micelles and free drugs. Binding and uptake of HB/FA-PEG-PLA micelles by SKOV3 cells were also observed in vivo after ip injection of folate-targeted micelles in tumor-bearing ascitic ovarian cancer animals. Drug levels in ascitic tumor tissues were increased 20-fold (P < .001), which underscored the effect of a regional therapy approach with folate targeting. Furthermore, the HB-loaded micelles were mainly distributed in kidney and liver (the main clearance organs) in biodistribution. These results showed that our newly developed PDT photosensitizer HB/FA-PEG-PLA micelles have a high drug-loading capacity, good biocompatibility, controlled drug release, and enhanced targeting and antitumor effect, which is a potential approach to future targeting ovarian cancer therapy.