EM-based segmentation-driven color standardization of digitized histopathology

Deep learning for digital pathology image analysis: A comprehensive tutorial with selected use cases

Skin color offers a strong cue for efficient localization and tracking of human body parts in video sequences for vision-based human-computer interaction. Color-based target localization could be achieved by analyzing segmented skin color regions. However, one of the challenges of color-based target tracking is that color distributions would change in different lighting conditions such that fixed color models would be inadequate to capture nonstationary color distributions over time. Meanwhile, using a fixed skin color model trained by the data of a specific person would probably not work well for other people. Although some work has been done on adaptive color models, this problem still needs further studies. We present our investigation of color-based image segmentation and nonstationary color-based target tracking, by studying two different representations for color distributions. We propose the structure adaptive self-organizing map (SASOM) neural network that serves as a new color model. Our experiments show that such a representation is powerful for efficient image segmentation. Then, we formulate the nonstationary color tracking problem as a model transduction problem, the solution of which offers a way to adapt and transduce color classifiers in nonstationary color distributions. To fulfill model transduction, we propose two algorithms, the SASOM transduction and the discriminant expectation-maximization (EM), based on the SASOM color model and the Gaussian mixture color model, respectively. Our extensive experiments on the task of real-time face/hand localization show that these two algorithms can successfully handle some difficulties in nonstationary color tracking. We also implemented a real-time face/hand localization system based on such algorithms for vision-based human-computer interaction.

The goal of this project was to develop two age appropriate atlases (neonatal and one year old) that account for the rapid growth and maturational changes that occur during early development. Tissue maps from this age group were initially created by manually correcting the resulting tissue maps after applying an expectation maximization (EM) algorithm and an adult atlas to pediatric subjects. The EM algorithm classified each voxel into one of ten possible tissue types including several subcortical structures. This was followed by a novel level set segmentation designed to improve differentiation between distal cortical gray matter and white matter. To minimize the req uired manual corrections, the adult atlas was registered to the pediatric scans using high -dimensional, symmetric image normalization (SyN) registration. The subject images were then mapped to an age specific atlas space, again using SyN registration, and the resulting transformation applied to the manually corrected tissue maps. The individual maps were averaged in the age specific atlas space and blurred to generate the age appropriate anatomical priors. The resulting anatomical priors were then used by the EM algorithm to re-segment the initial training set as well as an independent testing set. The results from the adult and age-specific anatomical priors were compared to the manually corrected results. The age appropriate atlas provided superior results as compared to the adult atlas. The image analysis pipeline used in this work was built using the open source software package BRAINSTools.

Objectives: The research aims to develop the segmentation model to identify the deformity in the medical images as accurately as possible and plan for better medical treatment. The study is extended to identify the disease before its appearance in the human body through human aura images to support aura imaging in medical diagnosis. Methods: The study used a brain image from the UCI data set and Aura images from the Biowell data set to identify the disease. The segmentation model Bivariate Gaussian Mixture Model (B.G.M.M) was developed. Model parameters are derived using the Expectation Maximization (E.M) Algorithm. The Grasshopper optimization Algorithm (G.O.A) extracts optimal features from the images. The chosen feature is fed as input to the classification model B.G.M.M. Segmentation accuracy is measured using the quality metrics. Findings: The developed approach shows 97% accuracy in identifying the damaged tissues in MRI images and high-intensity energy zones in the aura images, indicating the potential for deformities. Novelty: This study significantly contributes to the field by offering novel solutions for precise and comprehensive image analysis in medical and aura imaging contexts. Keywords: G.O.A, segmentation, G.M.M, E.M, quality metrics, deformity identification, Hue and saturation

Classification of different perfusion compartments in the brain is important to the profound analysis of brain perfusion. We presents a method based on a mixture of multivariate Gaussians (MoMG) and the expectation-maximization (EM) algorithm initialized by the results of hierarchical clustering (HC) on the whitened data to automatically dissect various perfusion compartments from dynamic susceptibility contrast (DSC) MR images so that each compartment comprises pixels of similar signal-time curves. Monte Carlo simulations have been designed and executed to assess the performance of the proposed method under various signal-to-noise ratios (SNRs). The results of simulations showed that using EM initialized by HC on whitened data produce the best accuracy of segmentation. Five healthy volunteers participated in this study for the validation of this method. The averaged ratios of gray matter to white matter for relative cerebral blood volume (rCBV), relative cerebral blood flow (rCBF) and mean transition time (MTT) derived from 5 normal subjects were 2.196±0.097, 2.259±0.119, and 0.968±0.023 which are in good agreement with those reported in the literature. The proposed method can subserve the diagnosis and assessment of various diseases involving the changes of cerebral blood distribution.

BackgroundEfforts are underway to eliminate trachoma as a public health problem by 2020. Programmatic guidelines are based on clinical signs that correlate poorly with Chlamydia trachomatis (Ct) infection in post-treatment and low-endemicity settings. Age-specific seroprevalence of anti Ct Pgp3 antibodies has been proposed as an alternative indicator of the need for intervention. To standardise the use of these tools, it is necessary to develop an analytical approach that performs reproducibly both within and between studies.MethodologyDried blood spots were collected in 2014 from children aged 1–9 years in Laos (n = 952) and Uganda (n = 2700) and from people aged 1–90 years in The Gambia (n = 1868). Anti-Pgp3 antibodies were detected by ELISA. A number of visual and statistical analytical approaches for defining serological status were compared.Principal FindingsSeroprevalence was estimated at 11.3% (Laos), 13.4% (Uganda) and 29.3% (The Gambia) by visual inspection of the inflection point. The expectation-maximisation algorithm estimated seroprevalence at 10.4% (Laos), 24.3% (Uganda) and 29.3% (The Gambia). Finite mixture model estimates were 15.6% (Laos), 17.1% (Uganda) and 26.2% (The Gambia). Receiver operating characteristic (ROC) curve analysis using a threshold calibrated against external reference specimens estimated the seroprevalence at 6.7% (Laos), 6.8% (Uganda) and 20.9% (The Gambia) when the threshold was set to optimise Youden’s J index. The ROC curve analysis was found to estimate seroprevalence at lower levels than estimates based on thresholds established using internal reference data. Thresholds defined using internal reference threshold methods did not vary substantially between population samples.ConclusionsInternally calibrated approaches to threshold specification are reproducible and consistent and thus have advantages over methods that require external calibrators. We propose that future serological analyses in trachoma use a finite mixture model or expectation-maximisation algorithm as a means of setting the threshold for ELISA data. This will facilitate standardisation and harmonisation between studies and eliminate the need to establish and maintain a global calibration standard.

This paper introduces a uniform statistical framework for both 3-D and 2-D object recognition using intensity images as input data. The theoretical part provides a mathematical tool for stochastic modeling. The algorithmic part introduces methods for automatic model generation, localization, and recognition of objects. 2-D images are used for learning the statistical appearance of 3-D objectss both the depth information and the matching between image and model features are missing for model generation. The implied incomplete data estimation problem is solved by the Expectation Maximization algorithm. This leads to a novel class of algorithms for automatic model generation from projections. The estimation of pose parameters corresponds to a non-linear maximum likelihood estimation problem which is solved by a global optimization procedure. Classification is done by the Bayesian decision rule. This work includes the experimental evaluation of the various facets of the presented approach. An empirical evaluation of learning algorithms and the comparison of different pose estimation algorithms show the feasibility of the proposed probabilistic framework.

1 - Congruence of deep learning in biomedical engineering: future prospects and challenges

Digital pathology incorporates the acquisition, management, sharing and interpretation of pathology information in a digital environment. The field of digital pathology is currently regarded as one of the most promising avenues of diagnostic medicine. Many computer-aided detection and diagnostic algorithms has been developed to assist pathologists in their daily clinical routine, with varying degree of success. These include cell detection and counting, tissue classification and cancer grading, among others. Deep learning, or more specifically, deep convolutional neural network, is a machine learning algorithm that has also gained a lot of attention recently due to their ability to achieve state-of-the-art accuracy. In this paper we have constructed and expanded the deep model network to localize tumor regions in histology images of breast carcinoma. We proposed our own deep convolutional neural network with lesser hardware requirement using 64×64×3 input patch. Our proposed method is able to provide reliable tumor region localization, visually and objectively, based on very limited training dataset.

Colorectal cancer (CRC) remains one of the leading causes of cancer-related mortality worldwide. Biomarkers such as microsatellite instability (MSI) play a pivotal role in guiding treatment decisions, particularly in the context of immunotherapy. However, conventional MSI testing methods, including PCR and sequencing, are often time-consuming and costly. To address this, we propose a transformer- based deep learning model for histopathological image analysis, specifically aimed at classifying tissue patches into nine categories to support MSI biomarker prediction. The model was trained on a publicly available CRC dataset from Zenodo, which includes annotated tissue patches categorized into nine classes: ADI, BACK, DEB, LYM, MUC, MUS, NORM, STR, and TUM. Patch features of shape (7×7×1024) were extracted using pretrained embeddings. A transformer encoder, followed by fully connected layers, was implemented using PyTorch. The model was trained with cross-entropy loss and optimized with Adam. Performance was evaluated using accuracy and confusion matrices. The transformer-based model achieved an overall classification accuracy of 96% on the test set. Notably, high precision and recall were observed for key classes such as TUM (tumor) and LYM (lymphocytes). Most misclassifications occurred between STR and DEB, which exhibit morphological similarities. Compared to conventional CNN-based approaches, the transformer model demonstrated superior generalization and interpretability, benefiting from its ability to model global dependencies through self-attention mechanisms. This study highlights the potential of transformer architectures for accurate and scalable tissue classification in digital pathology. The results confirm their applicability as a foundation for future biomarker prediction tasks, including MSI detection. Future work will focus on extending this framework for direct biomarker inference and validating its performance on larger, multi-institutional datasets.

Segmentation of MR images into tissue classes is often required in clinical applications. But automatic segmentation and tissue classification can be hardly satisfactory because of the overlapped distribution of intensities between tissues. This paper describes a new method called adaptive segmentation that uses knowledge of tissue properties and intensity inhomogeneities to correct and segment multi-spectral MR images. Use of the expectation-maximization (EM) algorithm leads to more accurate results.

In this paper we propose a new segmentation algorithm which combines patch-based information with edge cues under a probabilistic framework. We use a mixture of multiple Gaussians for building the statistical model with color and spatial features, and we incorporate edge information based on texture, color and brightness differences into the EM algorithm. We evaluate our results qualitatively and quantitatively on a large data-set of natural images and compare our results to other state-of-the-art methods.

Automated segmentation and classification of brain stroke using expectation-maximization and random forest classifier

The scarcity of large histopathological datasets can be problematic for Deep Learning in medical imaging and digital pathology. However, transfer Learning has been shown to be promising for the effective training of classifiers on smaller datasets. ImageNet is a popular dataset that is commonly used for transfer learning in various domains. The features extracted from the ImageNet dataset are generalizable and can be applied to alternative tasks and datasets. Deep Learning typically requires a vast amount of data for training, however, in our study we interrogated two datasets with patches extracted from only 30 whole slide images (WSIs) and 60 WSIs respectively. As a consequence, we decided to extract features and feed them into separate classifier models such as a fully connected softmax layer, Support Vector Machines (SVM) and Logistic Regression. This study demonstrated that for the small dataset, the best pretrained feature extractor was DenseNet201, whereas the best model for training was a fully connected softmax layer with a reported accuracy of 88.20% and an average f1-score of 0.881. For the larger dataset size, the best feature extractor was InceptionResNetV2 where the highest accuracy and f1-score of 90.60% and 0.908 was produced when classified using a fully connected softmax layer. All models, apart from ResNet50 demonstrated an improvement in performance when pretraining using ImageNet for bottleneck feature extraction.

On Digital Mammogram Segmentation and Microcalcification Detection Using Multiresolution Wavelet Analysis