Kubilay Muhammed Sunnetci

Background/Objectives: Anemia is a complex hematological disorder with various subtypes, and its accurate diagnosis is of critical importance for clinical decision-making processes. Traditional machine learning algorithms and standard deep learning models often struggle to model the nonlinear and typically imbalanced nature of clinical tabular data, which negatively impacts the classification performance and reliability of diagnostic systems. Methods: This study proposed ALSS-Net, a novel deep learning architecture that combines the State-Space Model (SSM), Liquid Time Constant (LTC), and Adaptive Gates mechanisms, with the aim of classifying difficult-to-diagnose types of anemia with high accuracy, robustness, and statistical reliability. Results: Experimental results showed that the Adaptive Liquid State-Space Networks (ALSS-Net) architecture achieved an accuracy of 92.97% on the validation set. Furthermore, on the test dataset, this model obtained an accuracy of 89.13% and an F1 score of 0.8513, outperforming five of the six baseline models by a statistically significant margin (McNemar’s test, p < 0.001); the difference with respect to the strongest baseline (MLP) did not reach statistical significance. The ablation studies indicated that the proposed dynamic time constant and adaptive gating mechanisms contributed to the separation of the difficult classes, with adaptive gating accounting for the larger part of the improvement. Conclusions: The architecture also produced better-calibrated probability outputs than the compared models, attaining the lowest Brier score and Expected Calibration Error, and degraded more gracefully under increasing feature noise. These findings were obtained on a single, relatively small open-access dataset without external clinical validation; ALSS-Net is therefore presented as a candidate framework for the future development of clinical decision support systems rather than as a validated diagnostic tool.

Camera networks now cover almost every public space, but the footage they generate has long outpaced what any human team can watch in real time. That gap is exactly where automated violence detection earns its place. The hard part is ambiguity—a hug or a bit of play-fighting can look a lot like a real attack, and systems that get this wrong flood operators with false alarms. Since violence lives in motion as much as in appearance, we build GoogLeNet-based Inflated 3D (I3D) two-stream classifiers that read space and time together. Three variants are trained and validated on the Real Life Violence Situations (RLVS) dataset (2000 videos, 80/20 split), each differing in temporal depth and resolution. The strongest, I3D Network 3, reaches 98.75% in-domain validation accuracy, with an F1 score of 98.74% and a Matthews Correlation Coefficient of 97.51%, values that should be read as upper bounds because of the file-level split and the validation-based checkpoint selection. Numbers on the home dataset only tell part of the story, so we push every model onto the independent RWF-2000 set, untouched during training. Even under this strict zero-shot transfer, I3D Network 3 holds 79.75% accuracy, although it is seen that this single value hides a strongly asymmetric error profile. The recall of the violence class decreases to 65.70%, in other words 343 of 1000 violent clips are missed, while the recall of the peaceful class remains at 93.80%, and McNemar and paired t-tests confirm the gaps between networks are real, rather than noise. For this reason, this result is evaluated as partial generalization. The only published values obtained under the same RLVS to RWF-2000 zero-shot protocol are 68.76% and 74.68%, so the margin is real but narrow. To open the black box, 3D Grad-CAM produces separate spatial and motion explanations that show where and when the model sees violence. Finally, the best model is embedded in ViolenceNet, a lightweight interface that keeps a human in the loop. Considering the missed-incident rate measured under domain shift, ViolenceNet is positioned as an assisted-review and triage tool, rather than as an autonomous alarm system.

Hemoglobin plays a critical role in diagnosing various medical conditions, including infections, trauma, hemolytic disorders, and Mediterranean anemia, which is particularly prevalent in Mediterranean populations. Conventional measurement methods require blood sampling and laboratory analysis, which are often time-consuming and impractical during emergency situations with limited medical infrastructure. Although portable oximeters enable non-invasive hemoglobin estimation, they still require physical contact, posing limitations for individuals with circulatory or dermatological conditions. Additionally, reliance on disposable probes increases operational costs. This study presents a non-contact and automated approach for estimating total hemoglobin levels from facial video data using three-dimensional regression models. A dataset was compiled from 279 volunteers, with synchronized acquisition of facial video and hemoglobin values using a commercial pulse oximeter. After preprocessing, the dataset was divided into training, validation, and test subsets. Three 3D convolutional regression models, including 3D CNN, channel attention-enhanced 3D CNN, and residual 3D CNN, were trained, and the most successful model was implemented in a graphical interface. Among these, the residual model achieved the most favorable performance on the test set, yielding an RMSE of 1.06, an MAE of 0.85, and a Pearson correlation coefficient of 0.73. This study offers a novel contribution by enabling contactless hemoglobin estimation from facial video using 3D CNN-based regression techniques.

Objectives: It is aimed at magnetic resonance imaging (MRI)-based differential diagnosis of parotid gland tumors (PGTs) using deep learning. Methods: This study included 117 PGTs obtained from 113 patients. T2-w, T1-w, contrast-enhanced T1-w, Diffusion Weighted Imaging-b0, Diffusion Weighted Imaging-b2000 (DWI-2000), and apparent diffusion coefficient sequences of these patients were used in the study. We implemented four different classification models, and we categorized the images as benign-malignant, pleomorphic adenoma (PA)-Warthin, Warthin-malignant, and all classes (mucoepidermoid carcinoma-other benign-other malignant-PA-Warthin). We constructed classification for each sequence separately using the ResNet18 architecture, with the dataset split into 80% for training and 20% for validation. Results: The most successful model in this study, achieving an accuracy of 95.37% and an F1-score of 94.74% in classifying malignant-Warthin images in T1-w sequences, also demonstrated the highest accuracy among all models evaluated. For the classification of benign-malignant and the differentiation across All classes, the highest accuracies were achieved with the T2 sequence at 93.75% and 86.67%, respectively. In the differentiation of PA-Warthin, T1-weighted and DWI-b0 sequences demonstrated the highest performance, both with an accuracy of 90.36%. Conclusion: The deep networks proposed in the study supported MRI-based differential diagnosis of PGTs with high accuracy, and the user-friendly software classified images with high accuracy in about 10 seconds.

Bladder cancer, often asymptomatic in the early stages, is a type of cancer where early detection is crucial. Herein, endoscopic images are meticulously evaluated by experts, and sometimes even by different disciplines, to identify tissue types. It is believed that the time spent by experts can be utilized for patient treatment with the creation of a computer-aided decision support system. For this purpose, in this study, it is evaluated that the performances of three models proposed using the bladder tissue dataset. The first model is a convolutional neural network (CNN)-based deep learning (DL) network, and the second is a model named hybrid cnn-machine learning (ML) or DL + ML, which involves classifying deep features obtained from a CNN-based network with ML. The last one, and the one that achieved the best performance metrics, is a vision transformer (ViT) architecture. Furthermore, a graphical user interface (GUI) is provided for an accessible decision support system. As a result, accuracy and F1 score values for DL, DL + ML, and ViT models are 0.9086–0.8971–0.9257 and 0.8884–0.8496–0.8931, respectively.