A Preliminary Study of Deep-Learning Algorithm for Analyzing Multiplex Immunofluorescence Biomarkers in Body Fluid Cytology Specimens

INTRODUCTION

Multiplex biomarker analysis of cytological body fluid specimens is often used to assist cytologists in distiguishing metastatic cancer cells from reactive mesothelial cells. However, evaluating biomarker expression visually may be challenging, especially when the cells of interest are scant. Deep-learning algorithms (DLAs) may be able to assist cytologists in analyzing multiple biomarker expression at the single cell level in the multiplex fluorescence imaging (MFI) setting. This preliminary study was performed to test the feasibility of using DLAs to identify immunofluorescence-stained metastatic adenocarcinoma cells in body fluid cytology samples.

METHODS

A DLA was developed to analyze MFI-stained cells in body fluid cytological samples. A total of 41 pleural fluid samples, comprising of 20 positives and 21 negatives, were retrospectively collected. Multiplex immunofluorescence labeling for MOC31, BerEP4, and calretinin, were performed on cell block sections, and results were analyzed by manual analysis (manual MFI) and DLA analysis (MFI-DLA) independently.

RESULTS

All cases with positive original cytological diagnoses showed positive results either by manual MFI or MFI-DLA, but 2 of the 14 (14.3%) original cytologically negative cases had rare cells with positive MOC31 and/or BerEP4 staining in addition to calretinin. Manual MFI analysis and MFI-DLA showed 100% concordance.

CONCLUSION

MFI combined with DLA provides a potential tool to assist in cytological diagnosis of metastatic malignancy in body fluid samples. Larger studies are warranted to test the clinical validity of the approach.

A deep-learning algorithm (AIFORIA) for classification of hematopoietic cells in bone marrow aspirate smears based on nine cell classes-a feasible approach for routine screening?

Bone marrow cytology plays a key role for the diagnosis and classification of hematological disease and is often the first step in the acute setting of unclear cytopenia. AI applications represent a powerful tool in digital image analysis and can improve the diagnostic workflow and accuracy. The aim of this study was to develop an algorithm for the automated detection and classification of hematopoietic cells in digitized bone marrow aspirate smears for potential implementation in the clinical laboratory. The AIFORIA create platform (Aiforia Technologies, Plc, Helsinki, Finland) was used to develop a convolutional neural network algorithm based on nine cell classes. Digitized bone marrow aspirate smears from normal hospital controls were used for AI training. External validation was performed on separate data sets. Automated cell classification was assessed in whole-slide images (WSI) and regions of interest (ROI). A total of 1950 single-cell annotations were applied for AI training with a final total class error of 0.15% with 99.9% precision and sensitivity (FI-score 99.2%). External validation showed an overall precision and sensitivity of 96% and 97% and a F1-score of 96%. Automated cell classification correlated highly across ROI with variable correlation to WSI. The average execution time for classifying 500 hematopoietic cells was < 1 s and ≤ 260 s for WSI. A cloud-based, deep-learning algorithm for automated detection and classification of hematopoietic cells in bone marrow aspirate smears is a very useful, reliable, and rapid screening tool in combination with cytomorphology.

Establishment of a corneal ulcer prognostic model based on machine learning

Corneal infection is a major public health concern worldwide and the most common cause of unilateral corneal blindness. Toxic effects of different microorganisms, such as bacteria and fungi, worsen keratitis leading to corneal perforation even with optimal drug treatment. The cornea forms the main refractive surface of the eye. Diseases affecting the cornea can cause severe visual impairment. Therefore, it is crucial to analyze the risk of corneal perforation and visual impairment in corneal ulcer patients for making early treatment strategies. The modeling of a fully automated prognostic model system was performed in two parts. In the first part, the dataset contained 4973 slit lamp images of corneal ulcer patients in three centers. A deep learning model was developed and tested for segmenting and classifying five lesions (corneal ulcer, corneal scar, hypopyon, corneal descementocele, and corneal neovascularization) in the eyes of corneal ulcer patients. Further, hierarchical quantification was carried out based on policy rules. In the second part, the dataset included clinical data (name, gender, age, best corrected visual acuity, and type of corneal ulcer) of 240 patients with corneal ulcers and respective 1010 slit lamp images under two light sources (natural light and cobalt blue light). The slit lamp images were then quantified hierarchically according to the policy rules developed in the first part of the modeling. Combining the above clinical data, the features were used to build the final prognostic model system for corneal ulcer perforation outcome and visual impairment using machine learning algorithms such as XGBoost, LightGBM. The ROC curve area (AUC value) evaluated the model's performance. For segmentation of the five lesions, the accuracy rates of hypopyon, descemetocele, corneal ulcer under blue light, and corneal neovascularization were 96.86, 91.64, 90.51, and 93.97, respectively. For the corneal scar lesion classification, the accuracy rate of the final model was 69.76. The XGBoost model performed the best in predicting the 1-month prognosis of patients, with an AUC of 0.81 (95% CI 0.63-1.00) for ulcer perforation and an AUC of 0.77 (95% CI 0.63-0.91) for visual impairment. In predicting the 3-month prognosis of patients, the XGBoost model received the best AUC of 0.97 (95% CI 0.92-1.00) for ulcer perforation, while the LightGBM model achieved the best performance with an AUC of 0.98 (95% CI 0.94-1.00) for visual impairment.

A global profiling strategy for identification of the total constituents in Chinese herbal medicine based on online comprehensive two-dimensional liquid chromatography-quadrupole time-of-flight mass spectrometry combined with intelligentized chemical classification guidance

A comprehensive strategy for effective identification of total constituents in Chinese patent medicine has been advanced applying full scan-preferred parent ions capture-static and active exclusion (FS-PIC-SAE) acquisition coupled with intelligent deep-learning supported mass defect filter (MDF) process, with Naoxintong capsule (NXT) as a case. Online comprehensive two-dimensional liquid chromatography (2DLC) coupled with Q-TOF-MS/MS system was established for obtaining the excellent separation and detection performance of total components, which could exhibit excellent peak capacity with 1052 and orthogonality with 0.69. In addition, a total of 901 unknown compounds could be classified into nine chemical classes rapidly and effectively, based on the intelligent deep-learning algorithm supported MDF model with 96.4% accuracy. Consequently, 276 compounds were successfully identified from NXT, especially including 44 flavonoids, 27 phenolic acids, 25 fatty acids, 17 saponins, 21 phthalocyanines, 20 triterpenes, 10 monoterpenes, 13 diterpenoid ketones, 14 amino acids, and others. It is concluded that the proposed program is an effective and practical strategy enabling the in-depth chemical profiling of complex herbal and biological samples.