Istanbul Technical University
Ulusal Üniversitelerarası Açık Erişim Sistemi - İstanbul Teknik ÜniversitesiNot a member yet
67356 research outputs found
Sort by
Annotation-efficient, patch-based, explainable deep learning using curriculum method for breast cancer detection in screening mammography
No full textAbstract Objectives To develop an efficient deep learning (DL) model for breast cancer detection in mammograms, utilizing both weak (image-level) and strong (bounding boxes) annotations and providing explainable artificial intelligence (XAI) with gradient-weighted class activation mapping (Grad-CAM), assessed by the ground truth overlap ratio. Methods Three radiologists annotated a balanced dataset of 1976 mammograms (cancer-positive and -negative) from three centers. We developed a patch-based DL model using curriculum learning, progressively increasing patch sizes during training. The model was trained under varying levels of strong supervision (0%, 20%, 40%, and 100% of the dataset), resulting in baseline, curriculum 20, curriculum 40, and curriculum 100 models. Training for each model was repeated ten times, with results presented as mean ± standard deviation. Model performance was also tested on an external dataset of 4276 mammograms to assess generalizability. Results F1 scores for the baseline, curriculum 20, curriculum 40, and curriculum 100 models were 80.55 ± 0.88, 82.41 ± 0.47, 83.03 ± 0.31, and 83.95 ± 0.55, respectively, with ground truth overlap ratios of 60.26 ± 1.91, 62.13 ± 1.2, 62.26 ± 1.52, and 64.18 ± 1.37. In the external dataset, F1 scores were 74.65 ± 1.35, 77.77 ± 0.73, 78.23 ± 1.78, and 78.73 ± 1.25, respectively, maintaining a similar performance trend. Conclusion Training DL models with a curriculum method and a patch-based approach yields satisfactory performance and XAI, even with a limited set of densely annotated data, offering a promising avenue for deploying DL in large-scale mammography datasets. Critical relevance This study introduces a DL model for mammography-based breast cancer detection, utilizing curriculum learning with limited, strongly labeled data. It showcases performance gains and better explainability, addressing challenges of extensive dataset needs and DL’s “black-box” nature. Key Points Increasing numbers of mammograms for radiologists to interpret pose a logistical challenge. We trained a DL model leveraging curriculum learning with mixed annotations for mammography. The DL model outperformed the baseline model with image-level annotations using only 20% of the strong labels. The study addresses the challenge of requiring extensive datasets and strong supervision for DL efficacy. The model demonstrated improved explainability through Grad-CAM, verified by a higher ground truth overlap ratio. He proposed approach also yielded robust performance on external testing data. Graphical Abstracthttps://doi.org/10.1186/s13244-025-01922-whttps://pubmed.ncbi.nlm.nih.gov/40106066http://dx.doi.org/10.1186/s13244-025-01922-whttps://doaj.org/article/51d2d181de7c4b168e7aaf2feebf3d3
Time-Dependent Tap Density Modeling of Graphite Milled by Vibrating Disc Mill
No full textGraphite, which is a key anode material for LIB, needs to have a high tap density (dt) to reach a high volumetric energy density. Since dt is directly correlated with particle size, particle size distribution, and particle shape, it can usually be improved by optimized grinding. So, determining the ideal grinding time by modeling the change in dt over grinding time can yield substantial benefits like time, energy, and economy. However, the grinding time-dependent dt modeling of graphite has never been reported before. Therefore, in this study, the relationship between the measured dt values and grinding times of graphite particles by a vibrating disc mill (VDM) was investigated. Then, the empirical time-dependent dt models were established with high R2 values. The experimental and predicted dt values were found to be close to each other. Among all tested fitting models, the exponential model (dt = ae−bt) was found to be the best-fitting model, having the highest R2 and lowest error values. This approach provides guidance in the powder flow and processing of ground mineral materials, in the preparation processes of high-density graphite LIB anode material, as well as in graphite grinding in other mills in the industry, as well as in different electrode materials.https://doi.org/10.3390/min15040403https://hdl.handle.net/20.500.12418/3611
The Role of Ensemble Deep Learning for Building Extraction from VHR Imagery
No full textIn modern geographical applications, the demand for up-to-date and accurate building maps is increasing, driven by essential needs in sustainable urban planning, sprawl monitoring, natural hazard mitigation, crisis management, smart city initiatives, and the establishment of climate-resilient urban environments. The unregulated growth in urbanization and settlement patterns poses multifaceted challenges, including ecological imbalances, loss of arable land, and increasing risk of drought. Leveraging recent technologies in remote sensing and artificial intelligence, particularly in the fields of very high-resolution satellite imagery and aerial photography, presents promising solutions for rapidly acquiring precise building maps. This research aims to investigate the efficiency of an ensemble deep learning framework comprising DeepLabV3+, UNet++, Pix2pix, Feature Pyramid Network, and Pyramid Scene Parsing Network architectures for the semantic segmentation of buildings. By employing the Wuhan University Aerial Building Dataset, characterized by a spatial resolution of 0.3 meters, as the training and testing dataset, the study assesses the performance of the proposed ensemble model. The findings reveal notable accuracies, with intersection over union metrics reaching 90.22% for DeepLabV3+, 91.01% for UNet++, 83.50% for Pix2pix, 88.90% for FPN, 88.20% for PSPNet, and finally at 91.06% for the ensemble model. These results reveal the potential of integrating diverse deep learning architectures to enhance the precision of building semantic segmentation.https://doi.org/10.26833/ijeg.158779
Real-Time 3D Point Cloud Segmentation on Edge Devices
No full texthttps://doi.org/10.1109/siu66497.2025.1111231
Manipulation of defect state emission in Zn chalcogenide quantum dots and their effects on chlorophyll spectral response
No full textWater soluble Zn based quantum dots (QDs) are of interest due to their biocompatibility and less toxic features. They have been frequently used in studies related to biotechnology, especially in agriculture studies. However, to control the optical properties of Zn based QDs has still been a challenge. In this work, the defect state emission of ZnSe QDs was successfully controlled through two different routes; 1) By creating a sulfur rich outer region around the Se rich core 2) By changing the capping agent. Gradient alloyed ZnSeS QDs with Se rich core and S rich outer region were successfully synthesized with two different capping agents; N-Acetyl-L-Cysteine (NAC) and 3-Mercaptopropionic Acid (3-MPA). The contribution of emission originated from surface-defects almost disappeared in NAC capped ZnSeS QDs, with causing a significant increase in photoluminescence quantum yield. The interaction between Zn based QDs with chlorophyll molecules was also investigated. The absorption capacity of chlorophylls significantly enhanced upon interaction with 3-MPA capped ZnSeS QDs. Also, the spectral response of chlorophylls could be modulated through interaction with 3-MPA capped ZnSeS QDs, which could be manipulated by using ZnSeS QDs with different chemical composition. Our results indicated that ZnSeS QDs have potential to be used in agriculture, which could act as a modulator of light-harvesting capacity of chlorophylls. The ability to modulate chlorophyll spectral responses through QD interaction opens new possibilities for optimizing light utilization in photosynthetic organisms, thereby contributing to enhanced crop yields and more efficient use of light energy in natural and artificial ecosystems.https://doi.org/10.1016/j.saa.2024.125348https://pubmed.ncbi.nlm.nih.gov/39481170https://aperta.ulakbim.gov.tr/record/28265
Prediction of pore water pressure generation of liquefiable clean sands under cyclic shear loading through deep learning
No full texthttps://doi.org/10.1080/17486025.2025.249149
Exploring Human Brain Metabolism via Genome-Scale Metabolic Modeling with Highlights on Multiple Sclerosis
No full textCerebral dysfunctions give rise to a wide range of neurological diseases due to the structural and functional complexity of the human brain stemming from the interactive cellular metabolism of its specific cells, including neurons and glial cells. In parallel with advances in isolation and measurement technologies, genome-scale metabolic models (GEMs) have become a powerful tool in the studies of systems biology to provide critical insights into the understanding of sophisticated eukaryotic systems. In this study, brain cell-specific GEMs were reconstructed for neurons, astrocytes, microglia, oligodendrocytes, and oligodendrocyte precursor cells by integrating single-cell RNA-seq data and global Human1 via a task-driven integrative network inference for tissues (tINIT) algorithm. Then, intercellular reactions among neurons, astrocytes, microglia, and oligodendrocytes were added to generate a combined brain model, iHumanBrain2690. This brain network was used in the prediction of metabolic alterations in glucose, ketone bodies, oxygen change, and reporter metabolites. Glucose supplementation increased the subsystems' activities in glycolysis, and ketone bodies elevated those in the TCA cycle and oxidative phosphorylation. Reporter metabolite analysis identified L-carnitine and arachidonate as the top reporter metabolites in gray and white matter microglia in multiple sclerosis (MS), respectively. Carbamoyl-phosphate was found to be the top reporter metabolite in primary progressive MS. Taken together, single and integrated iHumanBrain2690 metabolic networks help us elucidate complex metabolism in brain physiology and homeostasis in health and disease.https://doi.org/10.1021/acschemneuro.5c00006https://pubmed.ncbi.nlm.nih.gov/40091499http://dx.doi.org/10.1021/acschemneuro.5c0000
Exploring the Role of microRNAs as Blood Biomarkers in Alzheimer’s Disease and Frontotemporal Dementia
No full textAlzheimer’s disease (AD) and frontotemporal dementia (FTD) are the most common forms of dementia globally. AD is characterized by the accumulation of amyloid-β (Aβ) plaques and hyperphosphorylated tau in the brain, leading to progressive memory loss and cognitive decline, significantly impairing daily life. In contrast, FTD is marked by selective degeneration of the frontal and/or temporal lobes, typically resulting in profound changes in personality and social behavior, speech disorders, and psychiatric symptoms. Numerous studies have found microRNAs (miRNAs)—small, non-coding RNA molecules that regulate gene expression post-transcriptionally—to be dysregulated in AD and FTD. As a result, miRNAs have emerged as promising novel biomarkers for these diseases. This review examines the current understanding of miRNAs in AD and FTD, emphasizing their potential as accessible, noninvasive biomarkers for diagnosing these prevalent neurodegenerative disorders.https://doi.org/10.3390/ijms26073399https://pubmed.ncbi.nlm.nih.gov/40244285http://dx.doi.org/10.3390/ijms2607339
Development of a phthalocyanine-based photosensitizer for enhanced photodynamic inactivation of Gram-negative and Gram-positive bacteria
No full textLight-triggered antibacterial activity can be achieved with appropriate photosensitizers (PS). Quaternized PS are attractive molecules for antimicrobial photodynamic therapy (aPDT) applications. In this study, we synthesized a water-soluble Zn(II) phthalocyanine derivative (Full-COOH-ZnPc) that contains four quaternized groups on the periphery of the macrocycle as well as carboxylic groups. The photochemical and photophysical properties of the compound were examined to determine its potential as a PS in aPDT applications. The antibacterial activity of Full-COOH-ZnPcwas tested against Escherichia coli, Klebsiella pneumoniae, Pseudomonas aeruginosa, Staphylococcus aureus, Enterococcus faecalis, and Bacillus cereus. Analyses showed species- and dose-dependent responses to Full-COOH-ZnPc treatment following photoactivation, likely due to differences in cell membrane structure and protective mechanisms such as capsules. The results also highlighted the strong antibacterial effects of ZnPc, especially against [Formula: see text]. aureus and [Formula: see text]. faecalis, with no detectable dark toxicity, emphasizing the aPDI potential of the Pc compound against Gram-positive bacteria. Further optimization of the compound’s parameters and structure could improve its antimicrobial effectiveness against other bacterial species.https://doi.org/10.1142/s108842462550083
Spatial Autocorrelation Analysis of CO and NO2 Related to Forest Fire Dynamics
No full textThe increasing frequency and severity of forest fires globally highlight the critical need to understand their environmental impacts. This study applies spatial autocorrelation techniques to analyze the dispersion patterns of carbon monoxide (CO) and nitrogen dioxide (NO2) emissions during the 2021 Manavgat forest fires in Türkiye, using Sentinel-5P satellite data. Univariate (UV) Global Moran’s I values indicated strong spatial autocorrelation for CO (0.84–0.93) and NO2 (0.90–0.94), while Bivariate (BV) Global Moran’s I (0.69–0.84) demonstrated significant spatial correlations between the two gases. UV Local Moran’s I analysis identified distinct UV High-High (UV-HH) and UV Low-Low (UV-LL) clusters, with CO concentrations exceeding 0.10000 mol/m2 and exhibiting wide dispersion, while NO2 concentrations, above 0.00020 mol/m2, remained localized near intense fire zones due to its shorter atmospheric lifetime. BV Local Moran’s I analysis revealed overlapping BV-HH (high CO, high NO2) and BV-LL (low CO, low NO2) clusters, influenced by topography and meteorological factors. These findings enhance the understanding of gas emission dynamics during forest fires and provide critical insights into the influence of environmental and combustion processes on pollutant dispersion.https://doi.org/10.3390/ijgi14020065https://doaj.org/article/0fb2d6e5235a4044bc22e5af59b65a6