Optical sensing-based cytological image analysis enhanced with artificial intelligence: a review

Optical sensing technology is essential in cytological image analysis, particularly for cell morphology studies. Advancements in optical microscopy, such as bright-field, fluorescence, and confocal microscopy, have improved image resolution and acquisition speed, aiding early disease diagnosis. However, as sample volume and data complexity grow, traditional manual analysis struggles with efficiency and accuracy. Artificial intelligence (AI) is rapidly becoming a main method for automatic medical image analysis due to its powerful learning ability and pattern processing capability. Since cytology contributes greatly to early disease diagnosis and pathological assessment, recent research has focused on training AI for better utilization in intelligent cytological image analysis and received remarkable success. To provide a systematic and comprehensive understanding of the latest research progress, this paper reviews and summarizes AI methods applied in this field, including a detailed discussion on feature extraction methods (e.g., extraction of morphological features and color features of cells), segmentation methods (e.g., nucleus and parasite segmentation), detection methods (e.g., blood cell detection), classification methods (e.g., benign vs. malignant cell classification) and future challenge. This is followed by an overview of the application and development of artificial intelligence in detecting specific diseases, using research samples as a guide. Finally, the various challenges facing the field of intelligent analysis of cytological images, including problems from the lack of data sets, method selection, and clinical applications, are discussed in detail, and solutions that have proven to be informative and more promising are provided. By summarizing the latest solutions, this paper hopes to provide new ideas to address the imbalances in the development of cytological AI and the effective development of new methods in similar fields.