Abstract Cytological analysis plays a crucial role in diagnosing malignancies in peritoneal and pleural effusions. However, manual examination is time-consuming and subjective. This study proposes a novel computer vision-based approach to automate malignancy detection in cytological images. We employed the VGG16 convolutional neural network for feature extraction, combined with a Random Forest classifier to distinguish between benign and malignant cells. Several experiments were conducted using 5-fold cross-validation on a publicly available dataset to develop the proposed approach. The results were then compared with those reported in previous studies to evaluate performance improvements. The proposed model achieved an accuracy of 99.03%, precision of 99.40%, specificity of 100%, recall of 97.67%, and F1-score of 98.50%, surpassing several existing methods. The high specificity, minimizing false positives, is particularly advantageous in clinical settings, as it reduces unnecessary invasive procedures, patient anxiety, and healthcare costs. This study demonstrates the effectiveness of deep learning and machine learning in cytological analysis, offering a robust and highly accurate tool for assisting pathologists in malignancy detection. The systematic review highlights advancements and challenges in the field, reinforcing the potential of automated cytological diagnosis as a highly accurate and efficient tool to support pathologists and reduce diagnostic workload in clinical practice. The proposed approach is intended to support cytopathologists in decision-making and sample triage, contributing to more efficient and objective diagnostic workflows rather than replacing expert clinical judgment.
Rocha et al. (Sun,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: