Ghasem Azemi

Background: Accurate histological grading of gliomas, distinguishing low-grade (LGG) from high-grade (HGG) lesions, remains a critical determinant of treatment planning and patient prognosis. This study systematically investigates the optimal combination of structural MRI sequences, pretrained convolutional neural network (CNN) architectures, and supervised classifiers for automated glioma grade classification. Methods: Deep features were extracted from four MRI sequences (T1, contrast-enhanced T1 (T1Gd), T2, and FLAIR) from the BraTS 2023 dataset using five pretrained CNNs: VGG16, ResNet50, DenseNet121, EfficientNetB0, and InceptionV3. Features were refined via LASSO selection and classified using Random Forest, Support Vector Machine, XGBoost, Gradient Boosting, k-Nearest Neighbors (KNNs), and a shallow deep neural network. A patient-level 80/20 training–test partition was employed, with five-fold cross-validation used within the training set for feature selection and hyperparameter optimization. A total of 120 modality–extractor–classifier configurations were benchmarked on the held-out test set. Results: InceptionV3-derived features paired with KNN classifiers consistently yielded superior performance. Based on a TOPSIS multi-criteria ranking integrating accuracy, precision, recall, F1-score, and AUC, the best-performing configuration combined T1Gd features with KNN (closeness coefficient = 0.971; accuracy = 0.975, precision = 0.997, recall = 0.948, F1-score = 0.969, AUC = 0.996). The second-ranked configuration, using T1 features with KNN (closeness coefficient = 0.962), achieved significantly higher accuracy, recall, and F1-score, despite its slightly lower composite ranking. Conclusion: Deep feature extraction using InceptionV3 from T1-weighted MRI, coupled with KNN classification, represents a promising and practical approach for slice-level glioma grading and merits further validation in prospective clinical cohorts.

This study presents a fractal-based framework for quantifying visual expertise using eye-gaze data. A longitudinal experiment was conducted across three consecutive semesters of the medical program at Macquarie University, where the eye movements of 13 medical students were recorded with a high-resolution eye tracker while they viewed pathological images from multiple imaging modalities. In addition, eye-tracking data from 13 consultant neurosurgeons were included as expert reference data to assess whether longitudinal changes in students’ approach mirror expert-like visual behavior. Three metrics were computed: the two- and three-dimensional fractal dimensions of eye-gaze to capture spatiotemporal complexity of the eye-gaze data, and a fractal-dimension–based correlation to quantify the relationship between gaze patterns and stimulus structure. These metrics were combined into a composite Fractal Eye-Gaze Expertise Index (FEI). The results show systematic reductions in fractal complexity across sessions, indicating increasingly structured and efficient gaze strategies, with trends approaching those of experts. Crucially, conventional fixation-duration–based measures failed to consistently capture this progression, unlike the proposed fractal metrics. Statistical analyses confirm significant session-related effects for all metrics, and receiver operating characteristic curve (ROC) analysis shows that the FEI effectively discriminates longitudinal differences in visual expertise. This work demonstrates that fractal-based eye-gaze modelling provides an objective and scalable approach to characterizing visual expertise. Graphical abstract

Data preprocessing is a critical step in the analysis of matrix-assisted laser desorption/ionization–time-of-flight mass spectrometry (MALDI-TOF MS) spectra for machine learning applications, typically involving steps such as spectra trimming, baseline correction, smoothing, transformation, and peak picking or spectral binning. While traditional approaches focus on protein/peptide peaks as features, this study explores a novel method of feature extraction by treating MALDI-TOF spectra as one-dimensional signal array further processed as time-series data. This study investigates the use of computational fractal-based analysis to assess the complexity of MALDI-TOF spectra. Fractal analysis, previously successful in glioblastoma diagnosis using magnetic resonance imaging, was applied here to proteomics data. By treating each MALDI spectrum as a time series and calculating its fractal dimension using various algorithms, machine learning models were trained to differentiate between glioblastoma patients and controls. We demonstrate that fractal dimensions are sufficient to obtain accurate models for glioblastoma diagnosis, despite still underperforming when compared to the traditional feature extraction method. We also show that fractals can be used as support features to increase model performance. This work highlights the potential and limitations of fractal analysis in proteomics, offering a new perspective for disease diagnosis and broadening the available computational tools for data analysis in mass spectrometry.

Eye-tracking has been utilized to characterize visual behavior in medical image visualization and interpretation, yet neurosurgeons remain underrepresented. Characterizing neurosurgery-specific visual expertise is important for understanding expert search strategies, informing training, and developing computational models. This study examined gaze behavior in naïve observers (N_p = 29), neurosurgery registrars (N_p = 16), and consultant neurosurgeons (N_p = 24), viewing normal (N_p = 20) and pathological (N_p = 19) brain MR images under a free-viewing paradigm. To capture expertise-related characteristics, we analyzed two features at each fixation location: (i) fixation duration, reflecting temporal allocation of visual attention, and (ii) three-dimensional fractal dimension (3DFD) around each fixation location, quantifying local structural complexity. To assess pathological-type effects, we grouped similar pathologies into five stimulus groups. Linear mixed-effects modelling revealed systematic expertise-related differences, with experts exhibiting longer fixation durations in pathological stimulus groups and pathology-type-dependent complexity sampling. Combined fixation duration and 3DFD features captured complementary aspects of visual expertise, improving Random Forest classifier’s accuracy (>93%) compared to individual features, for all five stimulus groups. These findings highlight neurosurgery-specific markers of visual expertise and demonstrate that combining behavioral and image-derived features could underpin computational models and training tools that emulate expert-level strategies in neurosurgical image interpretation. Future work should evaluate its applicability to other medical domains.

Introduction: Rhinomanometry, a reference measure for the nasal airway, is often considered a research tool with only weak-to-moderate correlations with patient symptoms. However, like lung spirometry curves offer information beyond forced expiratory volume (FEV), rhinomanometry curves (rhinograms) have characteristics beyond simple nasal resistance at 150 Pascals. This study explored the correlation between rhinogram curve features and patient-reported outcomes (PROMs), when compared with nasal airway resistance. Methods: A diagnostic cross-sectional study was conducted on patients from a rhinology clinic. PROMs collected included ordinal nasal obstruction and visual analogue scale (VAS) of the more obstructed side. Rhinomanometry curves underwent mathematical polynomial fitting to extract 835 features. The primary outcome was correlation using Spearman’s rho (ρ) comparing curve-derived features with nasal airway resistance at 150 Pascals. Machine learning was applied to the top 8 correlated features to generate an AI predictive model. Results: About 601 patients (mean age 45 ± 16 years, 45% female) were analysed. Curve-derived features (ρ = 0.305) correlated more than total NAR at 150 Pa (ρ = 0.222) with VAS. Similarly with ordinal nasal obstruction, curve-derived features correlated more (ρ = 0.230) than total NAR at 150 Pa (ρ = 0.112). The best performing AI prediction models achieved correlations of 0.133 (VAS) and 0.117 (nasal obstruction). Conclusion: This study offers a novel method for rhinogram analysis with curve-derived features for correlation and predictive modelling. Whilst correlation scores remain weak-moderate with PROMs, they outperform nasal airway resistance. Therefore, rhinograms produced from rhinomanometry may offer more clinical information than a simplistic numerical resistance testing.

Objective: Artificial Intelligence (AI) research needs to be clinician led; however, expertise typically lies outside their skill set. Collaborations exist but are often commercially driven. Free and open‐source computational algorithms and software expertise are required for meaningful clinically driven AI medical research. Deep learning algorithms automate segmenting regions of interest for analysis and clinical translation. Numerous studies have automatically segmented paranasal sinus computed tomography (CT) scans; however, openly accessible algorithms capturing the sinonasal cavity remain scarce. The purpose of this study was to validate and provide an open‐source segmentation algorithm for paranasal sinus CTs for the otolaryngology research community. Methods: A cross‐sectional comparative study was conducted with a deep learning algorithm, UNet++, modified for automatic segmentation of paranasal sinuses CTs and “ground‐truth” manual segmentations. A dataset of 100 paranasal sinuses scans was manually segmented, with an 80/20 training/testing split. The algorithm is available at https://github.com/rheadkaul/SinusSegment . Primary outcomes included the Dice similarity coefficient (DSC) score, Intersection over Union (IoU), Hausdorff distance (HD), sensitivity, specificity, and visual similarity grading. Results: Twenty scans representing 7300 slices were assessed. The mean DSC was 0.87 and IoU 0.80, with HD 33.61 mm. The mean sensitivity was 83.98% and specificity 99.81%. The median visual similarity grading score was 3 (good). There were no statistically significant differences in outcomes with normal or diseased paranasal sinus CTs. Conclusion: Automatic segmentation of CT paranasal sinuses yields good results when compared with manual segmentation. This study provides an open‐source segmentation algorithm as a foundation and gateway for more complex AI‐based analysis of large datasets. Level of Evidence: 3

Background/Objectives: Tumor interactions with their surrounding environment, particularly in the case of peritumoral edema, play a significant role in tumor behavior and progression. While most studies focus on the radiomic features of the tumor core, this work investigates whether peritumoral edema exhibits distinct radiomic fingerprints specific to glioma (GLI), meningioma (MEN), and metastasis (MET). By analyzing these patterns, we aim to deepen our understanding of the tumor microenvironment’s role in tumor development and progression. Methods: Radiomic features were extracted from peritumoral edema regions in T1-weighted (T1), post-gadolinium T1-weighted (T1-c), T2-weighted (T2), and T2 Fluid-Attenuated Inversion Recovery (T2-FLAIR) sequences. Three classification tasks using those features were then conducted: differentiating between Low-Grade Glioma (LGG) and High-Grade Glioma (HGG), distinguishing GLI from MET and MEN, and examining all four tumor types, i.e., LGG, HGG, MET, and MEN, to observe how tumor-specific signatures manifest in peritumoral edema. Model performance was assessed using balanced accuracy derived from 10-fold cross-validation. Results: The radiomic fingerprints specific to tumor types were more distinct in the peritumoral regions of T1-c images compared to other modalities. The best models, utilizing all features extracted from the peritumoral regions of T1-c images, achieved balanced accuracies of 0.86, 0.81, and 0.76 for the LGG-HGG, GLI-MET-MEN, and LGG-HGG-MET-MEN tasks, respectively. Conclusions: This study demonstrates that peritumoral edema, as characterized by radiomic features extracted from MRIs, contains fingerprints specific to tumor type, providing a non-invasive approach to understanding tumor-brain interactions. The results of this study hold the potential for predicting recurrence, distinguishing progression from pseudo-progression, and assessing treatment-induced changes, particularly in gliomas.