Pathomics and Deep Learning Classification of a Heterogeneous Fluorescence Histology Image Dataset

Abstract

Automated pathology image classification through modern machine learning (ML) techniques in quantitative microscopy is an emerging AI application area aiming to alleviate the increased workload of pathologists and improve diagnostic accuracy and consistency. However, there are very few efforts focusing on fluorescence histology image data, which is a challenging task, not least due to the variable imaging acquisition parameters in pooled data, which can diminish the performance of ML-based decision support tools. To this end, this study introduces a harmonization preprocessing protocol for image classification within a heterogeneous fluorescence dataset in terms of image acquisition parameters and presents two state-of-the-art feature-based approaches for differentiating three classes of nuclei labelled by an expert based on (a) pathomics analysis scoring an accuracy (ACC) up to 0.957 ± 0.105, and, (b) transfer learning model exhibiting ACC up-to 0.951 ± 0.05. The proposed analysis pipelines offer good differentiation performance in the examined fluorescence histology image dataset despite the heterogeneity due to the lack of a standardized image acquisition protocol.

AI-boosted CRISPR-Cas13a and total internal reflection fluorescence microscopy system for SARS-CoV-2 detection

“…Gou et al (Gou et al, 2019) developed an image processing method that uses a two-step strategy for each fluorescent image to exclude auto-fluorescence or dust. Ioannidis et al (Ioannidis et al, 2021) reported a normalized preprocessing protocol for processing heterogeneous fluorescence images in terms of image magnification. Zhang et al (Zhang and Zhao 2019) applied preprocessing methods such as size adjustment and data expansion to each fluorescence dataset and then used CapsNet to classify the subcellular localization of proteins with an accuracy of 93.08%.…”

Deep Radiotranscriptomics of Non-Small Cell Lung Carcinoma for Assessing Molecular and Histology Subtypes with a Data-Driven Analysis

“…Thus, the “off-the-shelf” TL was used in the proposed methodology because of the low number of available samples in the examined patient cohort and the unbalanced natural prevalence of the disease. Additionally, this methodology has been successfully integrated into many medical image classification tasks, such as interstitial lung disease [ 37 ], colonic polyps [ 38 ], breast cancer [ 39 ], breast density assessment [ 40 ] and brain neoplasms [ 41 ], and evaluated across multiple other histopathology datasets [ 36 ].…”

A survey on recent trends in deep learning for nucleus segmentation from histopathology images

“…For the Seven Cell Line Dataset, the measured values of F1 are 0.865; for the LNCaP Dataset, the measured values of F1 are 0.874. PPU-Net similarly Limitations: To determine the optimal approach in cell microscopy, specialized object delineation methods and unique data sets outside the purview of the work are required Year: 2021 Ioannidis et al ( 2021 ) proposed a Pathomics and Deep Learning methodology for Fluorescence Histology Images Features: Backbone: Not mentioned Loss: Not mentioned The proposed module calculated the area of each object using the label function from Mahot's library, and it used all of the Pyradiomics library's accessible features, including statistical and higher-order statistical texture features. In addition, on the training set, three labeled nuclei types were achieved using the PyMRR library and the Mutual Information Differences (MID) technique to recognize an eloquent feature group Comparison: Xception, Residual Network (ResNet), Visual Geometry Group (VGG), Mobile-Net, Dense-Net and Nas-Net Datasets: Cell images of one distinct organ (Breast) were taken from Fluorescence Dataset (Kromp et al 2020 ) Parameters: Area Under Curve (AUC) and Accuracy (ACC) Inference: The proposed model was better in terms of performance as compared to other models in spite of the heterogeneity owing to the nonexistence of standardized image acquisition protocol, thus proving it to be applicable as a generalized and robust model in different scenario.…”