Staining, magnification, and algorithmic conditions for highly accurate cell detection and cell classification by deep learning.

OBJECTIVES: Research into cytodiagnosis has seen an active exploration of cell detection and classification using deep learning models. We aimed to clarify the challenges of magnification, staining methods, and false positives in creating general purpose deep learning-based cytology models. METHODS: Using 11 types of human cancer cell lines, we prepared Papanicolaou- and May-Grünwald-Giemsa (MGG)-stained specimens. We created deep learning models with different cell types, staining, and magnifications from each cell image using the You Only Look Once, version 8 (YOLOv8) algorithm. Detection and classification rates were calculated to compare the models. RESULTS: The classification rates of all the created models were over 95.9%. The highest detection rates of the Papanicolaou and MGG models were 92.3% and 91.3%, respectively. The highest detection rates of the object detection and instance segmentation models, which were 11 cell types with Papanicolaou staining, were 94.6% and 91.7%, respectively. CONCLUSIONS: We believe that the artificial intelligence technology of YOLOv8 has sufficient performance for applications in screening and cell classification in clinical settings. Conducting research to demonstrate the efficacy of YOLOv8 artificial intelligence technology on clinical specimens is crucial for overcoming the unique challenges associated with cytology.

Effect of liquid-based cytology fixing solution on immunocytochemistry: Efficacy of antigen retrieval in cytologic specimens.

BACKGROUND: Immunocytochemistry (ICC) is an indispensable technique to improve diagnostic accuracy. ICC using liquid-based cytology (LBC)-fixed specimens has been reported. However, problems may arise if the samples are not fixed appropriately. We investigated the relationship between the LBC fixing solution and ICC and the usefulness of antigen retrieval (AR) in LBC specimens. METHODS: Specimens were prepared from five types of LBC-fixed samples using cell lines and the SurePath™ method. ICC was performed using 13 antibodies and analyzed by counting the number of positive cells in the immunocytochemically stained specimens. RESULTS: Insufficient reactivity was observed using ICC without heat-induced AR (HIAR) in nuclear antigens. The number of positive cells increased in ICC with HIAR. The percentage of positive cells was lower in CytoRich™ Blue samples for Ki-67 and in CytoRich™ Red and TACAS™ Ruby samples for estrogen receptor and p63 than in the other samples. For cytoplasmic antigens, the percentage of positive cells for no-HIAR treatment specimens was low in the three antibodies used. In cytokeratin 5/6, the number of positive cells increased in all LBC specimens with HIAR, and the percentage of positive cells in CytoRich™ Red and TACAS™ Ruby samples was significantly lower (p < .01). For cell membrane antigens, CytoRich™ Blue samples had a lower percentage of positive cells than the other LBC-fixed samples. CONCLUSION: The combination of detected antigen, used cells, and fixing solution may have different effects on immunoreactivity. ICC using LBC specimens is a useful technique, but the staining conditions should be examined before performing ICC.

Relationship between a deep learning model and liquid-based cytological processing techniques.

OBJECTIVE: Artificial intelligence (AI)-based cytopathology studies conducted using deep learning have enabled cell detection and classification. Liquid-based cytology (LBC) has facilitated the standardisation of specimen preparation; however, cytomorphology varies according to the LBC processing technique used. In this study, we elucidated the relationship between two LBC techniques and cell detection and classification using a deep learning model. METHODS: Cytological specimens were prepared using the ThinPrep and SurePath methods. The accuracy of cell detection and cell classification was examined using the one- and five-cell models, which were trained with one and five cell types, respectively. RESULTS: When the same LBC processing techniques were used for the training and detection preparations, the cell detection and classification rates were high. The model trained on ThinPrep preparations was more accurate than that trained on SurePath. When the preparation types used for training and detection were different, the accuracy of cell detection and classification was significantly reduced (P < 0.01). The model trained on both ThinPrep and SurePath preparations exhibited slightly reduced cell detection and classification rates but was highly accurate. CONCLUSIONS: For the two LBC processing techniques, cytomorphology varied according to cell type; this difference affects the accuracy of cell detection and classification by deep learning. Therefore, for highly accurate cell detection and classification using AI, the same processing technique must be used for both training and detection. Our assessment also suggests that a deep learning model should be constructed using specimens prepared via a variety of processing techniques to construct a globally applicable AI model.

Effect of Specimen Processing Technique on Cell Detection and Classification by Artificial Intelligence.

OBJECTIVES: Cytomorphology is known to differ depending on the processing technique, and these differences pose a problem for automated diagnosis using deep learning. We examined the as-yet unclarified relationship between cell detection or classification using artificial intelligence (AI) and the AutoSmear (Sakura Finetek Japan) and liquid-based cytology (LBC) processing techniques. METHODS: The "You Only Look Once" (YOLO), version 5x, algorithm was trained on the AutoSmear and LBC preparations of 4 cell lines: lung cancer (LC), cervical cancer (CC), malignant pleural mesothelioma (MM), and esophageal cancer (EC). Detection and classification rates were used to evaluate the accuracy of cell detection. RESULTS: When preparations of the same processing technique were used for training and detection in the 1-cell (1C) model, the AutoSmear model had a higher detection rate than the LBC model. When different processing techniques were used for training and detection, detection rates of LC and CC were significantly lower in the 4-cell (4C) model than in the 1C model, and those of MM and EC were approximately 10% lower in the 4C model. CONCLUSIONS: In AI-based cell detection and classification, attention should be paid to cells whose morphologies change significantly depending on the processing technique, further suggesting the creation of a training model.

Immunocytochemical Reactivity Comparison between Formalin-Fixed and Liquid-Based Cytology-Fixed Specimens.

INTRODUCTION: Liquid-based cytology (LBC)-fixed samples can be used for preparing multiple specimens of the same quality and for immunocytochemistry (ICC); however, LBC fixing solutions affect immunoreactivity. Therefore, in this study, we examined the effect of LBC fixing solutions on immunoreactivity. METHODS: Samples were cell lines, and specimens were prepared from cell blocks of 10% neutral buffered formalin (NBF)-fixed samples and the four types of LBC-fixed samples: PreservCyt®, CytoRich™ Red, CytoRich™ Blue, and TACAS™ Ruby, which were post-fixed with NBF. ICC was performed using 24 different antibodies, and immunocytochemically stained specimens were analyzed for the percentage of positive cells. RESULTS: Immunoreactivity differed according to the type of antigen detected. For nuclear antigens, the highest percentage of positive cells of Ki-67, WT-1, ER, and p63 was observed in the NBF-fixed samples, and the highest percentage of positive cells of p53, TTF-1, and PgR was observed in the TACAS™ Ruby samples. For cytoplasmic antigens, the percentage of positive cells of CK5/6, Vimentin, and IMP3 in LBC-fixed samples was higher than or similar to that in NBF-fixed samples. The percentage of positive cells of CEA was significantly lower in CytoRich™ Red and CytoRich™ Blue samples than in the NBF-fixed sample (p < 0.01). Among the cell membrane antigens, the percentage of positive cells of Ber-EP4, CD10, and D2-40 was the highest in NBF-fixed samples and significantly lower in CytoRich™ Red and CytoRich™ Blue samples than that in NBF-fixed samples (p < 0.01). The NBF-fixed and LBC-fixed samples showed no significant differences in the percentage of positive cells of CA125 and EMA. DISCUSSION/CONCLUSION: ICC using LBC-fixed samples showed the same immunoreactivity as NBF-fixed samples when performed on cell block specimens post-fixed with NBF. The percentage of positive cells increased or decreased based on the type of fixing solution depending on the amount of antigen in the cells. Further, the detection rate of ICC with LBC-fixed samples varied according to the type of antibody and the amount of antigen in the cells. Therefore, we propose that ICC using LBC-fixed samples, including detection methods, should be carefully performed.

Relationship between Liquid-Based Cytology Preservative Solutions and Artificial Intelligence: Liquid-Based Cytology Specimen Cell Detection Using YOLOv5 Deep Convolutional Neural Network.

INTRODUCTION: Deep learning is a subset of machine learning that has contributed to significant changes in feature extraction and image classification and is being actively researched and developed in the field of cytopathology. Liquid-based cytology (LBC) enables standardized cytological preparation and is also applied to artificial intelligence (AI) research, but cytological features differ depending on the LBC preservative solution types. In this study, the relationship between cell detection by AI and the type of preservative solution used was examined. METHODS: The specimens were prepared from five preservative solutions of LBC and stained using the Papanicolaou method. The YOLOv5 deep convolutional neural network algorithm was used to create a deep learning model for each specimen, and a BRCPT model from five specimens was also created. Each model was compared to the specimen types used for detection. RESULTS: Among the six models, a difference in the detection rate of approximately 25% was observed depending on the detected specimen, and within specimens, a difference in the detection rate of approximately 20% was observed depending on the model. The BRCPT model had little variation in the detection rate depending on the type of the detected specimen. CONCLUSIONS: The same cells were treated with different preservative solutions, the cytologic features were different, and AI clarified the difference in cytologic features depending on the type of solution. The type of preservative solution used for training and detection had an extreme influence on cell detection using AI. Although the accuracy of the deep learning model is important, it is necessary to understand that cell morphology differs depending on the type of preservative solution, which is a factor affecting the detection rate of AI.

Effects of application method on skin penetration of carboxyfluorescein incorporated in liposomes.

We investigated the skin penetration of liposomes under two different application conditions; occluded and large application amount (1 ml/cm(2)), and open and small application amount (10 mul/cm(2)). Liposomes containing fluorescence-labeled phospholipids or carboxyfluorescein (CF) were used. In application under occluded conditions, phospholipids showed no penetration, even in the stratum corneum (SC). CF penetration in the skin after application of liposome was no different that after application of CF solution. In contrast, phospholipids penetrated the skin, particularly the SC and hair follicles, under open conditions. CF in liposome showed enhanced penetration in the SC and epidermis, but not in the dermis. On observation of the drying process, CF recrystallized from solution, but this did not occur with CF incorporated into liposome. It is possible that crystallization of CF is prevented by encapsulation in liposome, or that penetration occurs more readily with liposome.