1.Differentiating Brain Abscesses from Cystic and Necrotic Tumors Using Magnetic Resonance Imaging Features
Narantungalag A ; Tergel ch ; Tovuudorj A ; Delgerdalai kh
Mongolian Journal of Health Sciences 2026;95(5):165-170
Background:
Differentiating brain abscesses from necrotic cystic tumors is critical for clinical management but remains challenging due to overlapping features on conventional MRI.
Aim:
To evaluate the diagnostic performance of diffusion-weighted and morphological MRI biomarkers in differentiating brain abscesses from necrotic cystic brain tumors.
Materials and Methods:
This retrospective study included 54 patients with ring-enhancing brain lesions (brain abscess, n=13; metastasis, n=18; glioblastoma, n=23). MRI protocols (T1W, T2W, DWI, SWI, and T1+CE) were analyzed for lesion morphology, diffusion characteristics, capsule thickness, and the “dual rim sign.” Group comparisons were performed using ANOVA, chi-square, and Fisher’s exact tests. Diagnostic performance and optimal thresholds were determined using ROC curve analysis (p<0.05).
Results:
1. Comparison of Diffusion Parameters: The core ADC ratio min was significantly lower in the brain abscess group than in the cystic-necrotic tumor group (0.812±0.228 vs 1.902±0.535 for glioblastoma, 1.590±0.394 for metastasis; p<0.001). The area under the curve (AUC) for core ADC ratio min was 0.977, indicating excellent diagnostic performance. 2. Capsule Characteristics: On SWI, the “dual rim sign” was detected in 92.3% of brain abscesses but was absent (0%) in all cystic-necrotic tumors (p<0.001). The abscess capsules were significantly thinner (mean 3.3 mm) and more uniform compared to the irregular, thicker walls of necrotic tumors. 3. Diagnostic Thresholds: The optimal cut-off value for the core ADC ratio min was 0.94 (sensitivity 92.3%, specificity 92.7%). For capsule thickness, the cut-off value was 3.3 mm with an AUC of 0.955.
Conclusion
Core ADC ratio and capsule thickness are reliable quantitative MRI biomarkers for differentiating brain abscesses from cystic-necrotic tumors. Integrating diffusion metrics with susceptibility-based and morphological features, such as the “dual rim sign,” significantly enhances diagnostic accuracy in clinical practice, facilitating non-invasive and timely diagnosis.
2.Application of Artificial Intelligence-Based Ultrasound in the Differentiation of Benign and Malignant Breast Nodules
Lai-Fu Han ; Shiirevnyamba A ; Tsakhim-Erdene Ts ; Bat-Amgalan B ; Delgerdalai Kh
Mongolian Journal of Health Sciences 2026;95(5):223-227
Background:
Early and accurate differentiation of benign and malignant breast nodules is crucial for determining appropriate therapeutic strategies. Recently, artificial intelligence (AI)-based diagnostic systems have been increasingly utilized to enhance the diagnostic performance of ultrasound imaging.
Aim:
To evaluate the diagnostic value of artificial intelligence-based ultrasound (AI-US) in differentiating benign and malignant breast nodules and to compare its performance with conventional ultrasound (US).
Materials and Methods:
This retrospective study included 101 patients with 103 breast nodules (58 malignant, 45 benign) confirmed by histopathological analysis (biopsy or surgery). Diagnostic performance was evaluated across three groups: 1) Conventional ultrasound (US), 2) AI system (AI group), and 3) Combined US and AI system (US+AI group). Sensitivity (SE), specificity (SP), accuracy (ACC), positive predictive value (PPV), negative predictive value (NPV), and the area under the ROC curve (AUC) were calculated and compared.
Results:
Of the 103 nodules, 56.3% were malignant and 43.7% were benign. The AI group demonstrated higher diagnostic metrics than the US group—SE (82.76%), SP (80.00%), ACC (81.55%), and AUC (0.814)—though the difference was not statistically significant (p>0.05). However, the combined US+AI group showed significant improvement, reaching an SE of 87.93%, SP of 86.67%, ACC of 87.38%, and an AUC of 0.874. These results were statistically superior to the conventional US group (p<0.05).
Conclusion
The AI-based ultrasound diagnostic system significantly improves the accuracy of differentiating breast nodules. This technology serves as a reliable decision-support tool, particularly for reducing diagnostic errors among less experienced physicians and minimizing unnecessary invasive biopsies.
3.Comparative Study of Brain White Matter Hyperintensity Burden in Adults with Type 2 Diabetes
Tserensugir A ; ; Oyuntugs B ; Gonshigsuren D ; Tuvshinjargal D ; Tovuudorj A ; Mungunbagana G ; Munkhbaatar D ; Delgerdalai Kh
Mongolian Journal of Health Sciences 2026;95(5):253-256
Background:
Type 2 diabetes mellitus (T2DM) is a chronic metabolic disorder associated with cognitive impairment and structural brain alterations. White matter hyperintensities (WMH) detected on magnetic resonance imaging (MRI) serve as critical neuroimaging markers of cerebral small vessel disease. However, research evaluating the WMH burden in the Mongolian population with T2DM using AI-based automated quantitative methods and comparing these findings with a control group remains limited. This research gap provides the rationale and necessity for the current study.
Aim:
To evaluate and compare the burden of brain white matter hyperintensities (WMH) in adults with type 2 diabetes mellitus (T2DM) versus a control group.
Material and Methods:
This analytical cross-sectional, retrospective comparative study included participants aged 40 and older who underwent brain MRI at the Mongolia-Japan Hospital of MNUMS between 2019 and 2025. The T2DM group consisted of patients with a confirmed diagnosis, while the control group included individuals without diabetes or prediabetes. After excluding cases with major structural abnormalities, poor image quality, failed automated processing, or incomplete clinical data, 257 out of an initial 283 participants were included in the final analysis (T2DM: 71, Control: 186). 3D T1-weighted MPRAGE and T2-FLAIR MRI sequences were processed using Neurophet AQUA software (version 3.1) to assess total, deep, and periventricular Fazekas scores. Age- and sex-adjusted ordinal logistic regression analysis was performed using SPSS 25.0, with p<0.05 considered statistically significant.
Results:
A total of 257 participants (T2DM=71, Control=186) were analyzed. There were significant differences between groups in sex distribution (p=0.046) and age categories (p<0.001). The T2DM group exhibited significantly higher total and deep Fazekas scores compared to the control group (total Fazekas: aOR 10.81, 95% CI 5.20-22.49; deep Fazekas: aOR 8.21, 95% CI 4.32-15.60; both p<0.001). Increasing age was associated with higher total (aOR 1.07, 95% CI 1.03-1.11, p=0.001) and deep (aOR 1.04, 95% CI 1.00-1.08, p=0.037) Fazekas scores. Additionally, female sex was associated with higher total (aOR 1.95, 95% CI 1.04–3.63, p=0.036) and deep (aOR 1.95, 95% CI 1.08–3.51, p=0.026) Fazekas score categories.
Conclusion
Assessment of WMH burden using AI-based automated quantitative MRI processing revealed that T2DM is significantly associated with higher total and deep Fazekas scores. This association remained significant after adjusting for age and sex. These findings suggest that AI-based MRI processing is a suitable method for the objective, quantitative, and standardized assessment of brain WMH burden associated with T2DM.
4.Evaluation of DenseNet121 decision thresholds and probability calibration on chest radiographs from Mongolian individuals
Bayasgalant D ; Orgil-Erdene E ; Oyunbold L ; Delgerdalai Kh
Mongolian Journal of Health Sciences 2026;97(7):106-110
Background:
Mongolia’s vast territory and dispersed population pose challenges to timely diagnostic care. Artificial intelligence may support chest radiograph interpretation and prioritization of suspicious findings for clinician review. Models trained on foreign datasets should therefore be evaluated on Mongolian data before clinical use.
Aim:
To externally validate a DenseNet121 chest radiograph AI model trained on the public CheXpert dataset using Mongolian clinical data, and to evaluate local adaptation strategies for clinical translation.
Materials and Methods:
A DenseNet121-based multi-label classifier was trained on CheXpert with ImageNet initialization, 320×320 RGB inputs, and 14 sigmoid outputs, including No Finding and Support Devices. The best checkpoint was selected by internal CheXpert validation and evaluated without fine-tuning on 671 Mongolian chest radiographs whose 14 labels were manually extracted by a radiologist from previous clinical reports. Discrimination was assessed using ROC-AUC, macro AUC, and micro AUC. Threshold-dependent performance was evaluated at the default 0.5 threshold and after label-specific Youden index optimization. Youden thresholds were selected and evaluated on the same 671 images, so improvements may be optimistic. Platt calibration used five-fold patient-grouped cross-validation for 12 eligible labels; each patient was evaluated using a calibrator fitted without their data. Brier score and expected calibration error (ECE) were assessed.
Results:
The mean 14-label AUC in the internal CheXpert validation set used for model selection was 0.8169. On the Mongolian external validation set, macro AUC was 0.7517 and micro AUC was 0.7762. Strong discrimination was observed for Support Devices (0.9190), No Finding (0.8986), Consolidation (0.8914), Pleural Effusion (0.8793), Edema (0.8425), Pneumothorax (0.8193), and Lung Opacity (0.7997). AUC decreased for Fracture, Atelectasis, Pleural Other, and Cardiomegaly, with the causes of these differences unestablished. At the default 0.5 threshold, macro sensitivity was low (0.2347) despite high macro accuracy (0.8139), indicating that direct transfer may miss positive findings. Local threshold tuning improved F1 scores for Consolidation from 0.006 to 0.794, Edema from 0.509 to 0.600, Lung Opacity from 0.590 to 0.662, and Pleural Effusion from 0.737 to 0.757. Platt scaling improved probability reliability for several labels without systematically increasing AUC.
Conclusion
A CheXpert-trained DenseNet121 model showed moderate transferability to Mongolian chest radiographs, with strong performance for several clinically important findings and lower performance for other categories. Threshold optimization and calibration are feasible first-stage adaptation strategies. Larger multi-site Mongolian datasets, consensus labels, independent calibration/test sets, and prospective validation are required before clinical deployment.
5.Results of AI-assisted evaluation of abnormal findings on chest radiographs
Tsakhim-Erdene Ts ; Erdenebal G ; Vaanchigsuren S ; Tugsjargal P ; Delgerdalai Kh ; Tuvshinjargal D ; Munkhbaatar D
Mongolian Journal of Health Sciences 2026;97(7):180-184
Background:
Chest radiography is a fundamental imaging modality used to diagnose acute and chronic pulmonary diseases. It is relatively inexpensive, rapid, and can be performed at the bedside. Pulmonary abnormalities identified on chest radiographs are generally classified into two main categories: diffuse and focal. Depending on lesion size and image quality, abnormalities can be detected with an estimated accuracy of 70–85%. However, radiologists may overlook findings due to several factors, including heavy workloads, emergency conditions, and the presence of small focal lesions measuring less than 1 cm. Previous studies evaluating the application of artificial intelligence (AI) to chest radiographs for detecting lung cancer, COVID-19, and pneumonia have reported that the sensitivity of diagnostic interpretations improves from 88% to 95% when radiologists and AI work together.
Aim:
To evaluate selected thoracic abnormalities on chest radiographs using artificial intelligence and compare the results with radiologists’ interpretations.
Materials and Methods:
This retrospective study was conducted in the Department of Radiology at the Mongolia–Japan Hospital of the Mongolian National University of Medical Sciences. A total of 845 chest radiographs obtained between September 10 and September 30, 2022, were reviewed. After applying the exclusion criteria, 721 chest radiographs were included. The interpretations provided by four radiologists were compared with the results generated by an AI system.
Results:
A total of 721 posteroanterior chest radiographs were included in the study, comprising 377 images from male patients and 344 from female patients. By age group, the images included 24 from patients aged 10–19 years, 114 from those aged 20–29 years, 150 from those aged 30–39 years, 147 from those aged 40–49 years, 97 from those aged 50–59 years, 96 from those aged 60–69 years, 58 from those aged 70–79 years, and 37 from those aged 80–89 years. Radiologists identified abnormalities in 331 (46%) radiographs and reported no abnormalities in 390 (54%). In comparison, the AI system classified 392 (54%) radiographs as abnormal and 329 (46%) as normal. Using the radiologists’ interpretations as the reference standard, the AI system demonstrated a specificity of 95.8% and a sensitivity of 89% for detecting abnormalities on chest radiographs.
Conclusion
The AI system demonstrated high sensitivity and specificity in detecting abnormalities on chest radiographs, suggesting its potential utility in clinical practice. Its application may be particularly beneficial in supporting the interpretation of chest radiographs used for preventive screening and early detection of pulmonary abnormalities.
6.Association between AI-based whole-brain volume measurements and radiology report–based atrophy grades (A Quantitative MRI Study in Adults with and Without Type 2 Diabetes)
Tserensugir A ; ; Oyuntugs B ; Gonshigsuren D ; Tuvshinjargal D ; Tovuudorj A ; Mungunbagana G ; Munkhbaatar D ; Delgerdalai Kh
Mongolian Journal of Health Sciences 2026;97(7):185-189
Background:
The Global Cortical Atrophy (GCA) scale is widely used for visual assessment of cortical atrophy on MRI but is subject to interobserver variability. Automated brain volumetry provides objective measurements; however, its association with visual atrophy grading remains understudied in Mongolia.
Aim:
To assess the association and agreement between radiology report–based atrophy grades and AI-derived whole-brain volume, and to compare brain volumes in adults with and without type 2 diabetes mellitus (T2DM).
Materials and Methods:
This retrospective, single-center study included 367 adults aged ≥40 years (T2DM, n=144; controls, n=223). Intracranial volume–adjusted whole-brain volumes were obtained from 3D T1-weighted MRI using Neurophet AQUA v3.1. Atrophy severity recorded in routine radiology reports was coded from 0 to 3. Welch’s ANOVA assessed differences across grades, and ROC analysis evaluated discrimination between adjacent grades. Classification agreement was assessed using stratified five-fold cross-validation and Cohen’s κ. Age- and sex-adjusted ANCOVA compared brain volumes between groups.
Results:
Mean whole-brain volume decreased across atrophy grades 0–3 (1122.12, 1068.50, 1034.76, and 968.33 mL; Welch’s F(3, 60.04)=207.30; P<0.001). ROC AUCs for adjacent grades were 0.892, 0.897, and 1.000, with corresponding thresholds of 1099.86, 1056.17, and 995.57 mL. Five-fold cross-validation yielded 78.7% exact agreement (289/367), unweighted κ=0.684, and quadratic-weighted κ=0.856. Adjusted whole-brain volume was 11.98 mL lower in T2DM than in controls (95% CI, −22.67 to −1.30; P=0.028), whereas the overall association between T2DM and atrophy grade was not statistically significant (P=0.059).
Conclusion
Higher report-based atrophy grades were associated with lower whole-brain volumes, with strong quadratic-weighted agreement between volumetric and report-based classifications. Adults with T2DM had lower adjusted brain volumes than controls. Automated volumetry may complement radiologists’ assessments by providing objective quantitative information.
Result Analysis
Print
Save
E-mail