Deep learning technology has been used in the medical field to produce devices for clinical practice. Deep learning methods in cytology offer the potential to enhance cancer screening while also providing quantitative, objective, and highly reproducible testing. However, constructing high-accuracy deep learning models necessitates a significant amount of manually labeled data, which takes time. To address this issue, we used the Noisy Student Training technique to create a binary classification deep learning model for cervical cytology screening, which reduces the quantity of labeled data necessary. We used 140 whole-slide images from liquid-based cytology specimens, 50 of which were low-grade squamous intraepithelial lesions, 50 were high-grade squamous intraepithelial lesions, and 40 were negative samples. We extracted 56,996 images from the slides and then used them to train and test the model. We trained the EfficientNet using 2,600 manually labeled images to generate additional pseudo labels for the unlabeled data and then self-trained it within a student-teacher framework. Based on the presence or absence of abnormal cells, the created model was used to classify the images as normal or abnormal. The Grad-CAM approach was used to visualize the image components that contributed to the classification. The model achieved an area under the curve of 0.908, accuracy of 0.873, and F1-score of 0.833 with our test data. We also explored the optimal confidence threshold score and optimal augmentation approaches for low-magnification images. Our model efficiently classified normal and abnormal images at low magnification with high reliability, making it a promising screening tool for cervical cytology.
A double-check process helps prevent errors and ensures quality control. However, it may lead to decreased personal accountability, reduced effort, and declining quality checks. Introducing an artificial intelligence (AI)-based system in such scenarios could effectively address the risk of oversights. This study introduces an innovative AI-integrated workflow for cervical cytology screening that substantially improves efficiency and reduces the burden on cytologists. The AI model prioritizes cases for review based on anomaly scores and streamlines the first screening process to approximately 10 s per case. The model enhances the identification of high-risk cases via detailed microscopic observation, high anomaly scores cases, and a targeted review of low-score cases. The workflow highlights its capability for rapid, accurate, and less labor-intensive evaluations, demonstrating the potential to transform cervical cancer screening. This study highlights the importance of AI in modern medical diagnostics, particularly in areas with a high demand for accuracy and efficiency.
Cervical cytology is a crucial method for detecting cancerous and precancerous lesions. However, traditional workflows rely heavily on manual microscopic observations by cytotechnologists, making the process time-consuming and labor-intensive. Although several artificial intelligence (AI)-assisted cytology systems have been developed, most approaches require whole slide images, which entails costly scanning equipment, extensive data storage, and additional processing time. These factors hinder real-time diagnosis and are often impractical in resource-limited settings. In this study, we developed cytology-all-in-one (CYTOLONE), a novel AI model designed to support cytotechnologists in cervical cytology. CYTOLONE was constructed using a model based on OpenAI's contrastive language-image pretraining framework and fine-tuned using a hierarchical labeling structure. This approach enabled the model to effectively learn the relationship between low-magnification images and cytologic features. By integrating the microscope directly with an Apple Silicon Mac and using an iPhone camera for image capture, CYTOLONE offers real-time evaluation, processing each image in under 0.5 seconds. The evaluation results demonstrated that CYTOLONE achieved superior classification accuracy compared with both contrastive language-image pretraining-ViT-B/16 and GynAIe-B16-10k. The model maintained high Anomaly detection accuracy (95.8%) and significantly improved accuracy in Malignancy (92.8%), Bethesda (61.5%), and Diagnosis (57.5%) categories. Furthermore, feature space visualization revealed clearer boundaries between diagnostic categories, reflecting CYTOLONE's improved performance. The proposed workflow seamlessly integrates traditional cytology practices, allowing cytotechnologists to receive AI support during real-time specimen observation. This innovative workflow eliminates the need for costly whole slide image scanners, improves diagnostic efficiency, and is well-suited for resource-limited environments. Our findings suggest that CYTOLONE can enhance the efficiency of cytotechnologists and improve diagnostic accuracy, offering a practical solution to the existing limitations in AI-assisted cytology systems.
Background/Objectives: Transthoracic echocardiography (TTE) is a non-invasive tool for real-time assessment of cardiac motion and blood flow. It is widely used in emergency and bedside settings as a Focus Cardiac Ultrasound (FoCUS) device. However, standardized training methods and adequate educational environments are limited. Methods: A TTE image assessment artificial intelligence (AI) system was developed in this study, focusing on probe positioning and image quality for non-supervised TTE practice. Results: The view classification model achieved a high F1-score of 0.956. The position evaluation model achieved F1-scores of 0.678, 0.864, and 0.831 for the parasternal long-axis, parasternal short-axis, and apical four-chamber views, respectively. The quality evaluation model achieved F1-scores of 0.674, 0.845, and 0.746. Combining the position and quality models improved the F1-score for the parasternal long-axis view to 0.714, showing the benefit of integrating views with lower baseline performance. Conclusions: This study presents a novel AI-based educational system that assesses probe position and image quality in TTE. The model was developed using a custom dataset of healthy young adults that reflects beginner-level training scenarios, including many suboptimal images similar to those commonly acquired by novices. The proposed framework, which integrates position and quality models, lays the groundwork for future AI-assisted ultrasound training, particularly in unsupervised or resource-limited settings.
Background: Cytology artificial intelligence (AI) systems often depend on whole‐slide images (WSIs) or high‐magnification cell images and are commonly optimized for a single specimen type. We developed CYTOLONE‐v2, a contrastive language‐image pretraining (CLIP)–based hierarchical classification framework for cervical and urinary cytology that is designed for individual‐field inference without WSI input. Methods: CYTOLONE‐v1.1, previously trained for cervical cytology, was further trained by low‐rank adaptation using WSI‐derived urinary cytology fields equivalent to ×10 magnification. The 90‐case source cohort was divided at the case level into training (72 cases; 360 fields) and validation (18 cases; 90 fields) subsets. Five diagnosis‐representative fields were manually selected from each case. Field‐level performance was evaluated in an independent urinary test cohort of 40 cases (20 Reactive change and 20 Urothelial carcinoma; 200 fields), and retention was assessed using an independent cervical cytology test set. Results: In urinary cytology, CYTOLONE‐v2 achieved field‐level accuracies of 64.0% for Anomaly and Malignancy and 72.5% for Diagnosis. At the Diagnosis level, sensitivity for urothelial carcinoma was 99.0% (95% confidence interval [CI], 96.7%–100%), whereas specificity against Reactive change was 46.0% (95% CI, 30.0%–62.0%). Cervical cytology accuracy was retained at the Anomaly level (94.8% vs. 94.7% with CYTOLONE‐v1.1), decreased modestly at the Malignancy level (91.2% vs. 93.5%), and decreased more at the Bethesda and Diagnosis levels. Conclusions: CYTOLONE‐v2 demonstrates the technical feasibility of extending a hierarchical cytology classification framework from cervical to urinary cytology. The findings are limited to selected WSI‐derived fields and a Reactive change‐versus‐Urothelial carcinoma test cohort; external, case‐level, direct‐microscopy and prospective workflow validation is required before clinical utility can be established.