Talshyn Sarsembayeva

Бұл жұмыс инсульттан кейінгі науқастарды оңалтуға арналған Microsoft Kinect v2 сенсоры мен Unity платформасына негізделген ойындық жүйені әзірлеу және зерттеуге арналған. Зерттеудің өзектілігі үй жағдайында моторлық функцияларды қалпына келтіруге мүмкіндік беретін, сонымен қатар мониторингтің жоғары дәлдігін сақтайтын қолжетімді телереабилитация құралдарын жасау қажеттілігімен байланысты. Мақалада қаңқалық трекинг деректерін өңдеу үшін стохастикалық модельдеу әдістері мен экспоненциалды жылжымалы орташа (EMA) тәсілін қолдану егжей-тегжейлі сипатталады, бұл пациенттің ерікті қозғалыстарын патологиялық «шулардан» (діріл, спастика) тиімді ажыратуға мүмкіндік береді. Терапияның негізгі құралы ретінде когнитивтік тежелу мен моторлық координацияны бір уақытта жаттықтыруға арналған қосарланған тапсырма (dual-task training) модулін біріктіретін «Ball Catch» ойын механикасы ұсынылады. Жүйенің негізгі инновациясы - орындалу табыстылығына (SuccessRate) сүйене отырып, жаттығу параметрлерін (объектілердің жылдамдығы мен өлшемін) нақты уақыт режимінде реттейтін және шамадан тыс шаршаудың алдын алатын адаптивті қиындық алгоритмі. Екі пациентке 10 күн бойы жүргізілген пилоттық тестілеу нәтижелері оң динамиканы көрсетті. Буындардағы қозғалыс амплитудасының (ROM) 37-53% артуы және қозғалыс симметриясы индексінің 1,2-1,3° жақсаруы. Алынған деректер жанаспайтын сенсорларды алгоритмдік сүзгілеумен үйлестіру прогресті клиникалық стандарттарға шамалас дәлдікпен бағалауға мүмкіндік беретінін және науқастардың бетпе-бет әрі қашықтан емдеу үдерісіне тартылуын арттыратынын дәлелдейді.

We present a reproducible, web-scale pipeline for building a Kazakh-language corpus from the national e-government portal. The system treats the website as a directed graph and performs breadth-first traversal to preserve section hierarchy. Static acquisition relies on robust HTTP requests and HTML parsing; for pages with dynamic widgets, we selectively enable a headless layer to render the final DOM prior to extraction. We define a minimal JSON schema aligned with downstream NLP needs (URL, category, titles, cleaned descriptions) and implement normalization (Unicode NFC/NFKC, transliteration repair for Kazakh, boilerplate removal) and fragment-level deduplication. To strengthen the scientific contribution, we formalize the crawling–extraction process as an optimization under resource constraints and propose field-level quality metrics (precision, recall, F1), coverage of categories, and completeness gains attributable to headless rendering. Our experimental protocol compares static parsing against a hybrid static+headless setup on multiple portal categories, reports field-wise effectiveness with confidence intervals, and analyzes dominant error sources (DOM drift, client-side rendering, code-switching). Ablation studies quantify the impact of normalization and duplication. We also outline ethical access (robots.txt compliance, throttling, conditional requests) and provide artifacts to ensure reproducibility (versioned scripts, schema validators, logging). We release open-source scripts, detailed runbooks, and a small, labeled benchmark to facilitate fair comparisons and independent replication across institutions. The resulting corpus targets low-resource Kazakh NLP and e-government analytics, supporting tasks such as classification, terminology normalization, named-entity recognition, and LLM adaptation. Overall, the proposed pipeline demonstrates that selective headless rendering combined with rigorous normalization is a practical and effective strategy for high-quality data acquisition in dynamically rendered public portals.

This study presents a methodology for the automated detection and quantification of pulmonary emphysema from low-dose chest computed tomography (CT) scans. As a morphological subtype of chronic obstructive pulmonary disease (COPD), emphysema can be accurately assessed on CT imaging. Our approach utilizes a pre-trained ResNet152 model to extract high-dimensional feature embeddings (2048 dimensions) from mid-lung patches. Patients were automatically categorized based on the percentage of low attenuation areas (LAA%) below –950 Hounsfield units (HU), a standard measure for emphysema severity. The extracted feature embeddings were subsequently analyzed using statistical methods and logistic regression to identify key discriminative and interpretable features. A logistic regression model, trained on the top 20 most salient features, achieved a high level of performance, with an Area Under the Curve (AUC) of 0.94 and an Average Precision (AP) of 0.87 on a balanced dataset of 90 subjects. Furthermore, the selected features exhibited a strong correlation with LAA%, demonstrating their utility for regression-based severity assessment. The findings confirm the viability of using pre-trained deep embeddings for transparent and reproducible emphysema screening. This method avoids the need for extensive end-to-end model retraining, making it highly adaptable for integration into existing clinical CT analysis workflows. Keywords: Emphysema, low-dose CT, deep learning, ResNet, feature embeddings, explainable AI, COPD.

Accurate and interpretable detection of pulmonary emphysema on low-dose computed tomography (LDCT) remains a critical challenge for large-scale screening and population health studies. This work proposes a quality-controlled and interpretable deep learning pipeline for emphysema assessment using ResNet-152 embeddings. The pipeline integrates automated lung segmentation, quality-control filtering, and extraction of 2048-dimensional embeddings from mid-lung patches, followed by analysis using logistic regression, LASSO, and recursive feature elimination (RFE). The embeddings are further fused with quantitative CT (QCT) markers, including %LAA, Perc15, and total lung volume (TLV), to enhance robustness and interpretability. Bootstrapped validation demonstrates strong diagnostic performance (ROC-AUC = 0.996, PR-AUC = 0.962, balanced accuracy = 0.931) with low computational cost. The proposed approach shows that ResNet embeddings pretrained on CT data can be effectively reused without retraining for emphysema characterization, providing a reproducible and explainable framework suitable as a research and screening-support framework for population-level LDCT analysis.

This study addresses the automated conversion of unstructured Kazakh journalistic text into structured tabular representations. Existing text-to-table approaches are commonly developed for high-resource languages, rely on fixed schemas, or require computationally expensive full-model fine-tuning. These limitations reduce their applicability to morphologically rich low-resource languages and heterogeneous news collections. We propose a morphology-aware dual-regime framework for Kazakh text-to-table generation. The framework extends a Text-Tuple-Table pipeline with semantic chunking based on cosine-similarity thresholds, thematic grouping using k-means clustering, anchor-based factual cue detection, and large language model-based table generation. It supports both a static regime with a predefined five-column schema and a dynamic regime in which table headers and structure are inferred from the source article. A training resource was constructed from 149,624 articles published by Egemen Qazaqstan between 2017 and 2025. It contains 35,000 labeled text–table pairs, including 5000 manually annotated pairs and 30,000 semi-automatically generated pairs. A Qwen3.5-4B model was adapted using Low-Rank Adaptation under single-GPU constraints. Evaluation on a 1,000-record human-validated benchmark shows an in-domain trade-off between factual richness and structural regularity. The dynamic regime achieved higher Coverage (0.718), Accuracy (0.761), and supplementary Journalistic Value (0.932), whereas the static regime achieved higher Compression (0.969) and Structure (0.989). Ablation results further indicate that mixed supervision, thematic clustering, and Kazakh-specific preprocessing contribute to factual extraction quality. The findings demonstrate that combining morphology-aware processing, dual-regime schema generation, and parameter-efficient adaptation provides a practical approach to structured extraction from Kazakh journalistic text. Publicly available implementation materials and pretrained adapters support further reproducibility-oriented research in Kazakh and other low-resource language settings.

This research is dedicated to the development of an intelligent biometric system based on the synergy of Internet of Things (IoT) technologies and Artificial Intelligence (AI). The primary goal of this research is to explore the possibilities of personal identification using two distinct biometric traits: facial images and the venous pattern of the palm. These methods are treated as independent approaches, each relying on unique anatomical features of the human body. This study analyzes state-of-the-art methods in computer vision and neural network architectures and presents experimental results related to the extraction and comparison of biometric features. For each biometric modality, specific approaches to data collection, preprocessing, and analysis are proposed. We frame optimization in practical terms: selecting an edge-suitable backbone (ResNet-50) and employing metric learning (Triplet Loss) to improve convergence and generalization while adapting the stack for edge IoT deployment (Dockerized FastAPI with JWT). This clarifies that “optimization” in our title refers to model selection, loss design, and deployment efficiency on constrained devices. Additionally, the system’s architectural principles are described, including the design of the web interface and server infrastructure. The proposed solution demonstrates the potential of intelligent biometric technologies in applications such as automated access control systems, educational institutions, smart buildings, and other areas where high reliability and resistance to spoofing are essential.

This study presents a rule-based hybrid pipeline for the automated extraction of structured metadata from PDF versions of Kazakh-language newspaper articles, focusing on the national newspaper Egemen Qazaqstan. The primary goal is to support the development of a machine-readable knowledge base for future use in training large language models (LLMs) and building an AI-powered assistant for data journalism in Kazakhstan. The pipeline integrates three open-source parsers – pdfminer.six, PyMuPDF, and pdfplumber – to extract key elements such as title, author, date, abstract, text, journal name, and category. To evaluate extraction quality, we compared the results of the automated parser against manually annotated reference files across three real-world issues of the newspaper. The evaluation employed three complementary metrics: Precision, Textual Semantic Similarity (TSS), and Holistic Precision, which jointly assess both exact and semantic matches. The experimental results show that most tags – especially structured fields like date, journal, and category – achieved perfect Holistic Precision (1.00), while more variable fields like title still scored above 0.85. The validated pipeline was then applied to the full corpus of 2,140 newspaper PDFs published between 2017 and March 2025, successfully converting 159,135 articles into structured JSON format. This enriched corpus serves as a foundational knowledge base for Kazakh-language AI systems in journalism and media analysis.

The accurate segmentation of lung regions in computed tomography (CT) scans is critical for the automated analysis of lung diseases such as chronic obstructive pulmonary disease (COPD) and COVID-19. This paper focuses on enhancing the accuracy of U-Net segmentation models through a robust preprocessing pipeline. The pipeline includes CT image normalization, binarization to extract lung regions, and morphological operations to remove artifacts. Additionally, the proposed method applies region-of-interest (ROI) filtering to isolate lung areas effectively. The dataset preprocessing significantly improves segmentation quality by providing clean and consistent input data for the U-Net model. Experimental results demonstrate that the Intersection over Union (IoU) and Dice coefficient exceeded 0.95 on training datasets. This work highlights the importance of preprocessing as a standalone step for optimizing deep learning-based medical image analysis.