Abdulkerim Çapar

Objective: Tumor-infiltrating lymphocytes (TILs) in breast cancer are one of the most important indicators of the immune response within the tumor microenvironment. They play a particularly significant prognostic and predictive role in triple-negative and HER2-positive subtypes. However, substantial inter-observer variability has been reported in TIL scoring among pathologists, which limits its reliability in clinical practice. The aim of this study was to evaluate the agreement between artificial intelligence (AI) models and pathologists in TIL scoring and to compare this agreement using different statistical approaches, thereby assessing the potential of AI integration into pathology practice. Materials and Methods: Digitized histopathological images of breast cancer cases were included in the study. Tumor regions annotated by pathologists were evaluated for both stromal TIL percentage and the proportion of stromal tumor area within each ROI, with assessments performed independently by three pathologists and two AI models. Agreement was assessed among pathologists, between pathologists and AI, and between AI models. Statistical analyses included intraclass correlation coefficient (ICC), Cohen’s and Fleiss’ kappa, correlation tests, and Bland–Altman analysis. In addition, categorical agreement was examined using different cut-off values. Results: Inter-pathologist agreement was high, with an ICC of 0.81. In contrast, the global agreement between pathologists and AI models was lower (ICC 0.41). Pairwise comparisons of pathologist–AI agreement yielded substantially lower ICC values (0.12–0.21), although this improved to 0.53 when three pathologists were assessed jointly with a single AI model. The strongest categorical agreement was observed with dichotomized TIL scores (≤10% vs. >10%), whereas multi-category classifications were associated with a marked reduction in kappa values. Spearman correlation coefficients between pathologists and AI models ranged from moderate to good (ρ = 0.48–0.81). Agreement between the two AI models themselves was moderate, with an ICC of 0.64. Conclusion: This study demonstrated that AI models were able to show a certain degree of correlation with pathologists in TIL scoring but remained insufficient in terms of absolute agreement. Differences were more pronounced at higher TIL scores. These findings highlight the need for standardized algorithms trained on larger and more diverse datasets before AI-based TIL scoring can be reliably implemented in clinical practice.

Background and Objective: Variational Autoencoders (VAEs) offer a powerful framework for unsupervised anomaly detection and data clustering, often surpassing traditional methods. A core strength of VAEs lies in their ability to model data distributions probabilistically, enabling robust identification of anomalies and clusters through reconstruction likelihood — a stochastic metric providing a principled alternative to deterministic error scores. Methods: We investigated how different VAE architectures, combining reconstruction likelihood with a learnable or data-driven prior, performed in a clustering task on a toy dataset such as MNIST. Results were verified using dimensionality reduction techniques like t-distributed Stochastic Neighbor Embedding (t-SNE) and Uniform Manifold Approximation and Projection (UMAP), alongside clustering algorithms such as k-means and Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN). Results: The VAE’s encoder inherently maps data points into a latent space exhibiting discernible cluster structure, as evidenced by alignment with ground truth labels. While dimensionality reduction techniques (both t-SNE and UMAP) facilitated the application of clustering algorithms (k-means and HDBSCAN), these methods were primarily used to visualize and interpret the latent space organization. Conclusions: This study demonstrates that VAEs effectively cluster data by implicitly encoding assignments in their latent representations. Determining cluster membership from encoder output, combined with reconstruction likelihood using semantic features, offers a principled approach for identifying typical samples and anomalies. Future research should focus on leveraging this inherent clustering capability of VAEs to enhance interpretability and facilitate clinical application.

Breast cancer, a widespread and life-threatening disease, necessitates precise diagnostic tools for improved patient outcomes. Tumor-Infiltrating Lymphocytes (TILs), reflective of the immune response against cancer cells, are pivotal in understanding breast cancer behavior. However, inter-observer variability in TILs scoring methods poses challenges to reliable assessments. This study introduces a novel and interpretable proof-of-principle framework comprising two innovative inter-observer agreement measures. The first method, Boundary-Weighted Fleiss’ Kappa (BWFK), addresses tissue segmentation predictions, focusing on mitigating disagreements along tissue boundaries. BWFK enhances the accuracy of stromal segmentation, providing a nuanced assessment of inter-observer agreement. The second proposed method, the Distance Based Cell Agreement Algorithm (DBCAA), eliminates the need for ground truth annotations in cell detection predictions. This innovative approach offers versatility across histopathological analyses, overcoming data availability challenges. Both methods were applied to assess inter-observer agreement using a clinical image dataset consisting of 25 images of invasive ductal breast carcinoma tissue, each annotated by four pathologists, serving as a proof-of-principle. Experimental investigations demonstrated that the BWFK method yielded gains of up to 32% compared to the standard Fleiss’ Kappa model. Furthermore, a procedure for conducting clinical validations of artificial intelligence (AI) based cell detection methods was elucidated. Thoroughly validated on a clinical dataset, the framework contributes to standardized, reliable, and interpretable inter-observer agreement assessments. This study is the first examination of inter-observer agreements in stromal segmentation and lymphocyte detection for the TILs scoring problem. The study emphasizes the potential impact of these measures in advancing histopathological image analysis, fostering consensus in TILs scoring, and ultimately improving breast cancer diagnostics and treatment planning. The source code and implementation guide for this study are accessible on our GitHub page, and the full clinical dataset is available for academic and research purposes on Kaggle.

Myelin is an essential component of the nervous system and myelin damage causes demyelination diseases. Myelin is a sheet of oligodendrocyte membrane wrapped around the neuronal axon. In the fluorescent images, experts manually identify myelin by co-localization of oligodendrocyte and axonal membranes that fit certain shape and size criteria. Because myelin wriggles along x-y-z axes, machine learning is ideal for its segmentation. However, machine-learning methods, especially convolutional neural networks (CNNs), require a high number of annotated images, which necessitate expert labor. To facilitate myelin annotation, we developed a workflow and software for myelin ground truth extraction from multi-spectral fluorescent images. Additionally, to the best of our knowledge, for the first time, a set of annotated myelin ground truths for machine learning applications were shared with the community.

Cervical squamous intra-epithelial lesions (SIL) are precursor cancer lesions and their diagnosis is important because patients have a chance to be cured before cancer develops. In the diagnosis of the disease, pathologists decide by considering the cell distribution from the basal to the upper membrane. The idea, inspired by the pathologists' point of view, is based on the fact that cell amounts differ in the basal, central, and upper regions of tissue according to the level of Cervical Intraep- ithelial Neoplasia (CIN). Therefore, histogram information can be used for tissue classification so that the model can be explainable. In this study, two different classification schemes are proposed to show that the local histogram is a useful feature for the classification of cervical tissues. The first classifier is Kullback Leibler divergence-based, and the second one is the classification of the histogram by combining the embedding feature vector from morpho- metric features. These algorithms have been tested on a public dataset. ¹ The method we propose in the study achieved an accuracy performance of 78 . 69% in a data set where morphology-based methods were 69 . 07% and Convo- lutional Neural Network (CNN) patch-based algorithms were 75 . 77%. The proposed statistical features are robust for tackling real-life problems as they operate independently of the lesions manifold.

Myelin is an essential component of the nervous system and myelin damage causes demyelination diseases. Myelin is a sheet of oligodendrocyte membrane wrapped around the neuronal axon. In the fluorescent images, experts manually identify myelin by co-localization of oligodendrocyte and axonal membranes that fit certain shape and size criteria. Because myelin wriggles along x-y-z axes, machine learning is ideal for its segmentation. However, machine-learning methods, especially convolutional neural networks (CNNs), require a high number of annotated images, which necessitate expert labor. To facilitate myelin annotation, we developed a workflow and software for myelin ground truth extraction from multi-spectral fluorescent images. Additionally, to the best of our knowledge, for the first time, a set of annotated myelin ground truths for machine learning applications were shared with the community.

Myelin is an essential component of the nervous system and myelin damage causes demyelination diseases. Myelin is a sheet of oligodendrocyte membrane wrapped around the neuronal axon. In the fluorescent images, experts manually identify myelin by co-localization of oligodendrocyte and axonal membranes that fit certain shape and size criteria. Because myelin wriggles along x-y-z axes, machine learning is ideal for its segmentation. However, machine-learning methods, especially convolutional neural networks (CNNs), require a high number of annotated images, which necessitate expert labor. To facilitate myelin annotation, we developed a workflow and software for myelin ground truth extraction from multi-spectral fluorescent images. Additionally, to the best of our knowledge, for the first time, a set of annotated myelin ground truths for machine learning applications were shared with the community.

Myelin is an essential component of the nervous system and myelin damage causes demyelination diseases. Myelin is a sheet of oligodendrocyte membrane wrapped around the neuronal axon. In the fluorescent images, experts manually identify myelin by co-localization of oligodendrocyte and axonal membranes that fit certain shape and size criteria. Because myelin wriggles along x-y-z axes, machine learning is ideal for its segmentation. However, machine-learning methods, especially convolutional neural networks (CNNs), require a high number of annotated images, which necessitates expert labor. To facilitate myelin annotation, we developed a workflow and a software for myelin ground truth extraction from multi-spectral fluorescent images. Additionally, we shared a set of myelin ground truths annotated using this workflow.

Oligodendrocytes wrap around the axons and form the myelin. Myelin facilitates rapid neural signal transmission. Any damage to myelin disrupts neuronal communication leading to neurological diseases such as multiple sclerosis (MS). There is no cure for MS. This is, in part, due to lack of an efficient method for myelin quantification during drug screening. In this study, an image analysis based myelin sheath detection method, DeepMQ, is developed. The method consists of a feature extraction step followed by a deep learning based binary classification module. The images, which were acquired on a confocal microscope contain three channels and multiple z-sections. Each channel represents either oligodendroyctes, neurons, or nuclei. During feature extraction, 26-neighbours of each voxel is mapped onto a 2D feature image. This image is, then, fed to the deep learning classifier, in order to detect myelin. Results indicate that 93.38% accuracy is achieved in a set of fluorescence microscope images of mouse stem cell-derived oligodendroyctes and neurons. To the best of authors’ knowledge, this is the first study utilizing image analysis along with machine learning techniques to quantify myelination.

Chromosome classification is the essential task of karyotyping analysis which is videly used for detecting genetic anomalities. Chromosomes' band profiles based feature vectors are used frequently in the classification of them. Extracting of band profiles specially in the bent parts of chromosome may be lossy. In this study, an special sub-pixel resolution based method for lossless exctraction of chromosomes' band profile is proposed. This method models the skeleton of chromosomes as piece-wise linear equations using normal lines to the chromosome boundary. Based on the obtained model, extracting of band profiles is done by forming the skeleton normal lines in the sub-pixel resolution. For evaluation the accuracy of band profile extraction, a measure defined as "Percent of Band Profile Loss" demonstrates the proposed method as an alternative method which results to lossless chromosome band profile extraction.