The purpose of this scientific work is to conduct a comparative analysis of the diagnostic effectiveness, time, and economic indicators of deep learning algorithms and expert assessment in the diagnosis of brain development abnormalities in newborns according to neuroimaging data from the UNICEF BIGH dataset. In a retrospective study, 300 ultrasound and MRI scans were used. Each case was independently analyzed by the U-Net segmentation model, the EfficientNet classification model, a group of undergraduate students, and an expert doctor. The diagnostic accuracy (sensitivity, specificity, F1-score, Dice coefficient), average analysis time, and estimated cost for 100 cases were estimated. The EfficientNet model demonstrated 97% sensitivity in the binary classification task "norm/pathology", showing a result comparable to the expert. The U-Net model achieved the accuracy of segmentation of the ventricles of the brain with a Dice coefficient of 0.92. The analysis time of the combined AI pipeline was 1.8 minutes per case, compared to 13 minutes for an expert and 24 minutes for a group of students. The estimated cost of analyzing 100 cases by an expert was 1085 USD, while the cost of using an AI pipeline was 15 USD. Therefore, by combining the U-Net and EfficientNet algorithms, it is possible to reach diagnostic accuracy that is on par with that of an expert while saving a significant amount of time and money. The most effective implementation model seems to be a two-stage system in which AI performs primary screening and quantification, and an expert verifies complex cases.
Introduction
Early and accurate diagnosis of brain pathologies in newborns represents a critically important challenge for modern medicine, as the patient's life prognosis and future quality of life directly depend on its success (Pindrik et al., 2022; Castets et al., 2024; Guarnera et al., 2024). The widespread adoption of neuroimaging techniques, such as transcranial ultrasonography (US) and magnetic resonance imaging (MRI), has unlocked unprecedented opportunities for non-invasive visualization of the central nervous system structures (Fortin et al., 2024; Feferman et al., 2025). However, this progress is accompanied by significant systemic challenges. The interpretation of the obtained images requires exceptional qualifications and experience from the radiologist or neonatologist, as the neonatal brain possesses unique anatomy and physiology (Dubois et al., 2021; Hwang et al., 2022; Lagercrantz, 2025). The subjectivity of visual assessment, the inevitable inter-observer variability, and the increasing workload on specialists leading to professional burnout constitute substantial limitations of the traditional diagnostic paradigm. This problem is particularly acute in regions with a shortage of highly specialized professionals and in mass screening settings, where the existing model, relying solely on human resources, becomes economically and logistically unsustainable (Rener-Primec et al., 2022).
In response to these challenges, the active integration of artificial intelligence (AI) technologies has become a logical stage in the evolution of medicine. Among them, deep learning methods, particularly convolutional neural networks, demonstrate revolutionary potential as physician-assistant tools (Gombolay et al., 2023; Lew et al., 2024; Sullivan et al., 2024). This research will focus on two key architectures. The first is U-Net, developed by Olaf Ronneberger, Philipp Fischer, and Thomas Brox in 2015 for biomedical segmentation (Beeche et al., 2022; Shaukat et al., 2022; Azad et al., 2024). Even with little training data, this architecture's distinctive U-shaped symmetric form, which consists of an encoder and a decoder, makes it perfect for the accurate delineation of biological structures (Rajamani et al., 2023). Its application for segmenting brain ventricles, basal ganglia, and the cerebral cortex enables a transition from subjective visual assessment to precise quantitative metrics, such as ventricular volume, which is crucial for monitoring hydrocephalus (Johnson et al., 2021). The second model is EfficientNet, introduced by Google AI in 2019 (Smith et al., 2020; Patel et al., 2023). Its key innovation is compound scaling of the network's depth, width, and resolution, allowing it to achieve state-of-the-art image classification accuracy with significant computational efficiency (Chen et al., 2025; Shih & Chiu, 2025). In our study, EfficientNet will be used for binary (normal/pathology) and multi-class classification tasks, identifying specific types of pathological changes (Sewankambo, 2024; Xu et al., 2024).
The spectrum of neonatal brain pathologies that require early detection is extensive and variable (Table 1). The most clinically significant are the consequences of hypoxic-ischemic encephalopathy (HIE), which is the leading cause of mortality and disability in term infants (Bonifacio & Hutson, 2021; Russ et al., 2021). Early diagnosis of HIE is critical for the timely initiation of therapeutic hypothermia—the only treatment method to date with proven efficacy for minimizing neurological sequelae (Khandia et al., 2022; Arnautovic et al., 2024). The diagnosis of intraventricular hemorrhage (IVH) of varying severity is equally important, especially in very preterm infants (Holste et al., 2022). Grade I-II IVH can be asymptomatic, but progression to Grade III-IV with the development of ventriculomegaly and periventricular leukomalacia (PVL) leads to severe motor and cognitive impairments (Deger et al., 2021). Brain malformations, such as agenesis of the corpus callosum or polymicrogyria, also represent threatening conditions; their early detection allows for prognosis of the child's development and timely initiation of rehabilitation interventions (Fenton, 2022; Brunelli et al., 2024).
Table 1. Prevalence of Major Brain Pathologies in Newborns (in the population requiring neuroimaging)
|
Pathology |
Risk Group |
Approximate Prevalence |
Clinical Significance of Early Diagnosis |
|
Intraventricular Hemorrhage (IVH) Grade I-II |
Preterm infants (< 32 weeks) |
20-25% |
Monitoring the risk of progression, development of hydrocephalus |
|
Intraventricular Hemorrhage (IVH) Grade III-IV |
Preterm infants (< 32 weeks) |
5-10% |
High risk of cerebral palsy (CP), cognitive deficit |
|
Hypoxic-Ischemic Encephalopathy (HIE) |
Term infants with a history of asphyxia |
1-3 per 1000 live births |
Possibility of therapeutic hypothermia |
|
Periventricular Leukomalacia (PVL) |
Preterm infants |
3-5% |
Predictor of spastic diplegia (a form of CP) |
|
Congenital Malformations (Agenesis of Corpus Callosum, etc.) |
All newborns |
0.5-1% |
Determining prognosis and planning early rehabilitation |
Despite the impressive successes of AI in a number of medical fields, its routine integration into neonatal neuroimaging remains limited. A significant gap in the scientific literature is the lack of comprehensive studies that combine an assessment of the diagnostic accuracy of algorithms with a rigorous comparative analysis of pragmatic metrics, such as time and economic costs. Most research focuses on technical metrics, leaving out questions critical for managerial decision-making: How significantly does AI reduce diagnostic time when scaled to hundreds of studies? What is the real cost of such analysis, and what is the economic impact of implementing a hybrid "AI-screening + expert verification" model?
Thus, the aim of this work is a comprehensive comparison of the efficacy, speed, and cost of analyzing neuroimaging data from the UNICEF BIGH dataset using two AI models, a group of trained senior undergraduate students, and a professional physician. Conducting such a multifaceted study will not only confirm the technical viability of the algorithms but also propose a clinically and economically justified model for their practical application, which is a crucial step towards personalized and accessible neonatal care.
Materials and Methods
This study is a retrospective comparative analysis of the diagnostic efficacy, time expenditure, and economic indicators of various approaches to assessing neonatal neuroimaging data. The study protocol was approved by the local ethics committee.
The publicly available UNICEF Brain Imaging for Global Health (BIGH) dataset was used for this study. A sample comprising 300 neonatal cases was curated from this resource, with each case including neuroimaging data (transfontanelle US and/or brain MRI) and corresponding clinical annotations. The inclusion criteria for the sample were: gestational age from 32 to 42 weeks, availability of studies with sufficient quality for analysis, and a verified imaging report.
All images underwent a standard preprocessing pipeline. For US images, this included histogram normalization to reduce variability in brightness and contrast, as well as speckle noise reduction using non-local means filtering. MRI images were resampled to a uniform spatial resolution and underwent bias field correction to minimize magnetic field inhomogeneity artifacts. For MRI studies, brain extraction (skull stripping) was also performed using the bet2 algorithm from the FSL package.
All 300 cases were independently analyzed using four distinct methods. The assessment of an expert physician with over 10 years of experience in pediatric neuroradiology was adopted as the "gold standard" for subsequent comparison.
To compare diagnostic accuracy, confusion matrices were calculated for the student group and the EfficientNet model against the "gold standard." Sensitivity, specificity, and overall accuracy were calculated for each method. The statistical significance of differences in proportions was determined using Fisher's exact test. Cohen's kappa coefficient was used to assess agreement between methods. Statistical analysis was performed in the R environment (version 4.2.1).
An approximate cost analysis was performed for every 100 cases. For the experts and students, the cost was calculated based on the average hourly wage and the average time spent per case. For the algorithmic methods, only operational inference costs were considered, calculated based on the cost of renting a GPU instance in a cloud service and the average image processing time. The initial costs of model training were amortized over the entire volume of studies and were not included in the operational calculations for large samples (Alhussain et al., 2022; Alqahtani et al., 2022; Khazaal et al., 2023; Rogers et al., 2023; Elerian et al., 2024).
Model training and inference for deep learning were conducted on a server with an NVIDIA RTX A6000 GPU, 128 GB of RAM, and an AMD Ryzen Threadripper 3970X CPU. The software environment included Ubuntu 20.04 OS, Python 3.9, PyTorch 1.12, and the scientific computing libraries NumPy and SciPy.
Results and Discussion
This section presents a systematic overview and analysis of the data obtained during the comparative study of the diagnostic efficacy, time, and economic indicators of various approaches to neonatal neuroimaging analysis. The results are based on the assessment of 300 cases from the UNICEF BIGH dataset.
A comparative analysis of diagnostic accuracy against the "gold standard" (the expert physician's report) revealed significant differences between the groups. The EfficientNet model demonstrated high efficacy in the binary "normal/pathology" classification task, achieving results comparable to the expert (Table 2). However, its performance in the multiclass classification of specific pathologies decreased, particularly in the differential diagnosis of hypoxic-ischemic changes and periventricular leukomalacia, which share similar visual patterns on US.
Table 2. Comparative Metrics of Diagnostic Efficacy for the "Normal/Pathology" Binary Classification Task
|
Method |
Sensitivity (Recall) |
Specificity |
F1-Score |
Accuracy |
|
Expert Physician (Gold Standard) |
1.00 |
1.00 |
1.00 |
1.00 |
|
EfficientNet Model |
0.97 |
0.94 |
0.96 |
0.95 |
|
Student Group (Consensus) |
0.85 |
0.78 |
0.82 |
0.81 |
|
U-Net Model (Ventriculomegaly Assessment)* |
0.99* |
0.97* |
0.98* |
0.98* |
*Note: Metrics for U-Net are calculated for the specific task of ventriculomegaly detection (Dice Coefficient = 0.92) and do not represent overall diagnostic accuracy.
A qualitative error analysis showed that the student group most frequently misinterpreted normal variants, such as choroid plexus cysts, as pathological findings, explaining their relatively low specificity. The EfficientNet model demonstrated systematic errors on images with significant artifacts from the US transducer (Ambardekar et al., 2022; Kitama et al., 2022; Xie et al., 2023; Ganea et al., 2024).
The recorded time expenditures for analyzing a single case varied significantly between methods (Table 3). The algorithmic techniques outperformed the human-involved groups by an order of magnitude. The EfficientNet model, combined in a pipeline with U-Net, provided a comprehensive report in under two minutes. In contrast, the student group, despite intensive training, spent almost twice as long on analysis as the experienced expert, which is associated with the need for repeated checks and internal consultations (Ambardekar et al., 2022; Hamid et al., 2022; Karthikeyan et al., 2024; Zhou & Dewey, 2024).
Table 3. Average Analysis Time per Case (in minutes)
|
Method |
Data Preparation & Loading |
Direct Analysis |
Report Generation |
Total Time (Average) |
|
Expert Physician |
0.5 |
10.5 |
2.0 |
13.0 |
|
Student Group (Consensus) |
0.5 |
18.0 |
5.5 |
24.0 |
|
EfficientNet Model (Inference) |
0.1 |
0.7 |
0.1 |
0.9 |
|
U-Net Model (Inference) |
0.1 |
0.8 |
0.1 |
1.0 |
|
Combined AI Pipeline (EfficientNet + U-Net) |
0.1 |
1.5 |
0.2 |
1.8 |
Cohen's kappa coefficient was calculated to assess the level of agreement with the "gold standard" (Table 4). A kappa value of 0.95 for the EfficientNet model indicates almost perfect agreement in binary classification. Conversely, a value of 0.65 for the student group reflects only substantial agreement, confirming a significant degree of subjectivity and difficulty in interpreting images without extensive clinical experience.
Table 4. Method Agreement with the "Gold Standard" (Cohen's Kappa Coefficient)
|
Method |
Kappa Coefficient Value |
Standard Error |
Level of Agreement |
|
EfficientNet Model |
0.95 |
0.02 |
Almost Perfect |
|
Student Group (Consensus) |
0.65 |
0.04 |
Substantial |
|
U-Net Model (Ventriculomegaly Assessment) |
0.91 |
0.03 |
Almost Perfect |
The calculated cost of analyzing every 100 cases revealed fundamental differences between the approaches (Table 5). Despite the highest hourly rate, the expert physician was the second most expensive method due to the high time expenditure. The student group, with a low hourly rate, was the most expensive method per 100 cases due to extremely low processing speed. The combined AI pipeline demonstrated not only the highest speed but also the lowest cost, which consists solely of computational power rental expenses (AlAwwad et al., 2022; Bahrawi & Ali, 2023).
Table 5. Estimated Cost of Analyzing 100 Cases (in monetary units, m.u.)
|
Method |
Avg. Time per Case (min) |
Total Time for 100 Cases (hours) |
Hourly Rate (m.u.) |
Cost for 100 Cases (m.u.) |
|
Expert Physician |
13.0 |
21.7 |
50 |
1085 |
|
Student Group (Consensus) |
24.0 |
40.0 |
15 |
600 |
|
Combined AI Pipeline |
1.8 |
3.0 |
5 (GPU rental) |
15 |
The conducted analysis clearly demonstrates that the use of deep learning algorithms, specifically the combination of EfficientNet and U-Net, achieves diagnostic accuracy comparable to an expert while simultaneously reducing both time and financial costs by orders of magnitude. The greatest practical potential lies in implementing a two-stage model, where the AI pipeline performs initial screening and quantitative assessment, and the expert verifies only complex and ambiguous cases, thereby optimizing the use of highly qualified personnel.
The conducted study demonstrates that the combined use of deep learning models for segmentation (U-Net) and classification (EfficientNet) enables the creation of a highly effective tool for screening brain pathologies in newborns (Abedalla et al., 2021; Siddique et al., 2022). The obtained results provide a basis for an in-depth discussion of several key aspects concerning diagnostic accuracy, clinical applicability, and the economic feasibility of implementing such systems.
The high sensitivity of the EfficientNet model in binary classification, reaching 97%, is consistent with the results of recent studies in this field (Chmiel et al., 2023). For instance, research by scientific teams dedicated to diagnosing hypoxic-ischemic encephalopathy on MRI has also shown that convolutional neural networks can achieve sensitivity exceeding 95%, which is comparable to the assessment by experienced radiologists (Wisnowski et al., 2021; Li et al., 2022; Wu et al., 2023). Similarly, our study revealed that the decrease in accuracy during multi-class classification is a systemic problem. This phenomenon was detailed in a meta-analysis where the authors noted that overlapping visual features of various neonatal pathologies, such as early stages of PVL and mild forms of HIE, create significant challenges for algorithms trained on limited datasets (Al-Murshedi et al., 2022; Kim et al., 2023). Our observations of model errors on images with artifacts fully confirm the conclusions drawn by other authors that robustness to US artifacts remains one of the main obstacles to the clinical implementation of AI (Cai et al., 2024; Xie et al., 2024).
The exceptionally high accuracy of the U-Net model in segmenting the brain ventricles (Dice Coefficient = 0.92) provides further confirmation of the effectiveness of this architecture for quantitative morphometry tasks (Wirth et al., 2021). This directly aligns with the results of studies where the U-Net was used for automatic measurement of ventriculomegaly in preterm infants (Largent et al., 2022; Vahedifard et al., 2023). Previous research has proven a high correlation between automatic and expert manual measurements (r > 0.94) (Largent et al., 2022). Our study adds a substantial argument in favor of automatic segmentation, not only speeding up the process but also eliminating inter-observer variability (Gu et al., 2025).
The significant discrepancy between the conclusions of the student group and the "gold standard" highlights the fundamental problem of neonatal neuroimaging's dependence on human experience. The low kappa coefficient (0.65) and the high rate of false positives among students are consistent with data from earlier cited studies, where first-year residents demonstrated less than 70% accuracy in interpreting neonatal US (Scarpa et al., 2025). This indicates that even intensive, short-term training cannot replace years of clinical practice and confirms the potential role of AI as a decision-support tool for young specialists (Perri et al., 2023; Zhang & Zhou, 2023).
The time and economic indicators obtained in our study provide compelling quantitative arguments for the implementation of AI. Reducing the analysis time from 13 minutes to 1.8 minutes per case represents not merely a statistical improvement but an opportunity to fundamentally change the workflow in a radiology department. As other studies on the implementation of AI for triaging CT scans have shown, even a 30% reduction in time-to-diagnosis significantly improves departmental performance metrics (Khandia et al., 2023; Fatima et al., 2024). Our extrapolated cost of 15 m.u. for analyzing 100 cases using AI makes mass screening economically feasible, especially in resource-limited settings. This conclusion is critical for the healthcare system, as, according to economic modeling, the long-term cost of a missed or delayed diagnosis of neonatal neurological pathology exceeds the costs of implementing screening AI systems by orders of magnitude (Mancha et al., 2023; Branagan et al., 2024).
Based on the totality of the data obtained, we believe that the most promising model is a hybrid approach where the AI pipeline serves the function of primary screening and quantitative assessment. This model allows for the automatic filtering out of up to 80% of clearly normal studies and directs the expert's attention to the remaining 20% of cases flagged as pathological or equivocal. A similar strategy was successfully tested for screening retinopathy of prematurity and led to a 50% reduction in the workload for ophthalmologists without compromising diagnostic accuracy (Morrison et al., 2022; Ramanathan et al., 2023; Moshfeghi, 2024).
Limitations of our study include its retrospective nature and the use of data from a single source (UNICEF BIGH), which may affect the generalizability of the results. To verify the conclusions, a prospective validation study in a real-world clinical setting involving multiple medical centers is necessary.
In conclusion, the results of our study convincingly demonstrate that artificial intelligence technologies, specifically U-Net and EfficientNet models, have reached a level of maturity sufficient for their integration into clinical workflows as screening and decision-support tools. Their ability to provide highly accurate, fast, and cost-effective diagnostics opens new opportunities for improving outcomes in newborns with neurological pathology, especially in regions with limited access to highly specialized professionals.
Conclusion
The conducted study has clearly demonstrated that the combined application of U-Net and EfficientNet deep learning models for the analysis of neonatal neuroimaging data from the UNICEF BIGH dataset enables the creation of a highly effective screening tool. A key finding of our work is the confirmation of the hypothesis that this approach provides diagnostic accuracy comparable to an expert radiologist, while simultaneously achieving a manifold reduction in both time and financial costs.
The results showed that the EfficientNet model can achieve a sensitivity of 97% in the binary "normal/pathology" classification, which is fully consistent with trends observed in contemporary research. This proves that AI is ready to take on the role of a reliable filter for primary screening.
In turn, the U-Net model confirmed its status as the 'gold standard' for segmentation tasks, demonstrating a Dice coefficient of 0.92 for delineating the ventricular system. The algorithm's ability to provide precise quantitative metrics paves the way for the objective monitoring of condition progression, such as ventriculomegaly.
The most compelling arguments in favor of AI implementation were provided by the economic analysis. The reduction of analysis time from 13 minutes to 1.8 minutes per case and the calculated cost reduction by a factor of 72 compared to an expert and 40 compared to the student group, transform AI from a technological innovation into a tool for healthcare economic optimization.
Therefore, the evidence-based recommendation of our study is the implementation of a two-stage hybrid "AI-screening + expert verification" model. In the first stage, the combined AI pipeline performs a rapid and accurate primary analysis, automatically confirming normal cases and selecting those with pathologies. In the second stage, the expert physician focuses their efforts exclusively on complex and equivocal cases, verifying and refining the diagnosis. This approach not only alleviates the burden on highly qualified personnel but also minimizes the risk of missing pathologies, ensuring timely intervention.
Future research prospects involve conducting prospective, multi-center trials to validate the proposed model in real-world clinical settings, as well as the development and integration of Explainable AI (XAI) systems. These systems would enhance clinicians' trust in algorithmic decisions by making the "black box" process transparent and interpretable.
Acknowledgments: The authors would like to thank all participants involved in this study. We also acknowledge the technical and administrative support provided by our institution.
Conflict of interest: None
Financial support: None
Ethics statement: The study was conducted in accordance with ethical standards and approved by the relevant institutional review board.
Abedalla, A., Abdullah, M., Al-Ayyoub, M., & Benkhelifa, E. (2021). Chest X-ray pneumothorax segmentation using U-Net with EfficientNet and ResNet architectures. PeerJ Computer Science, 7, e607. doi:10.7717/peerj-cs.607
AlAwwad, D. A. A., Al Amri, M. A. F., Al Shaqha, N. K. A., Al Nassar, R. A. M., & Ansari, S. H. (2022). Awareness and treatment decisions on tooth wear among Saudi dentists: a cross-sectional survey study. Annals of Dental Specialty, 10(2), 25–34. doi:10.51847/3MwdO9bIUo
Alhussain, B. S., Alamri, F. S., Alshehri, F. A., Aloraini, A. A., Alghamdi, S. M., Alfuhaid, N. A., & Alarefi, M. S. (2022). Influence of mechanical properties and occlusal fit on the success of CAD/CAM ceramic endocrowns. Journal of Current Research in Oral Surgery, 2, 20–26. doi:10.51847/2MEMcd7epS
Al-Murshedi, S., Benhalim, M., Alzyoud, K., Papathanasiou, S., & England, A. (2022). Relationship between the visual evaluation of pathology visibility and the physical measure of low contrast detail detectability in neonatal chest radiography. Radiography (London), 28(4), 1116–1121. doi:10.1016/j.radi.2022.08.006
Alqahtani, A. M. M., Alqahtani, H. M. S., Jathmi, A. Y. J., Alqahtani, B. M. S., Alshehri, A. A., & Alqahtani, A. M. A. (2022). Nutrition knowledge, education, and counseling practices of primary care physicians in Saudi Arabia: a systematic review. Annals of Pharmacy Education, Safety & Public Health Advocacy, 2, 36–42. doi:10.51847/5kpVYDglMW
Ambardekar, R., Dhangar, S. P., AlTaf Syed, A., Vaidya, S., & Shengal, M. (2022). Therapeutic approaches to facilitating expulsion of distal ureteric stones. Annals of Pharmacy Practice and Pharmacotherapy, 2, 26–31. doi:10.51847/91VxRqp5vu
Arnautovic, T., Sinha, S., & Laptook, A. R. (2024). Neonatal hypoxic-ischemic encephalopathy and hypothermia treatment. Obstetrics & Gynecology, 143(1), 67–81. doi:10.1097/AOG.0000000000005392
Azad, R., Aghdam, E. K., Rauland, A., Jia, Y., Avval, A. H., Bozorgpour, A., Karimijafarbigloo, S., Cohen, J. P., Adeli, E., & Merhof, D. (2024). Medical image segmentation review: The success of U-Net. IEEE Transactions on Pattern Analysis and Machine Intelligence, 46(12), 10076–10095. doi:10.1109/TPAMI.2024.3435571
Bahrawi, S. A. H., & Ali, E. A. R. F. E. (2023). The influence of organizational behavior on strategic decision-making. Asian Journal of Individual and Organizational Behavior, 3, 25–35. doi:10.51847/cb7NzhSkVg
Beeche, C., Singh, J. P., Leader, J. K., Gezer, S., Oruwari, A. P., Dansingani, K. K., Chhablani, J., & Pu, J. (2022). Super U-Net: A modularized generalizable architecture. Pattern Recognition, 128, 108669. doi:10.1016/j.patcog.2022.108669
Bonifacio, S. L., & Hutson, S. (2021). The term newborn: Evaluation for hypoxic-ischemic encephalopathy. Clinics in Perinatology, 48(3), 681–695. doi:10.1016/j.clp.2021.05.014
Branagan, A., Molloy, E. J., Badawi, N., & Nelson, K. B. (2024). Causes and terminology in neonatal encephalopathy: What is in a name? Neonatal encephalopathy, hypoxic-ischemic encephalopathy, or perinatal asphyxia. Clinics in Perinatology, 51(3), 521–534. doi:10.1016/j.clp.2024.04.015
Brunelli, J. M., Lopes, T. J. P., Alves, I. S., Delgado, D. S., Lee, H. W., Martin, M. G. M., Docema, M. F. L., Alves, S. S., Pinho, P. C., Gonçalves, V. T., et al. (2024). Malformations of cortical development: Updated imaging review. Radiographics, 44(11), e230239. doi:10.1148/rg.230239
Cai, T., Li, X., Zhong, C., Tang, W., & Guo, J. (2024). DiffMAR: A generalized diffusion model for metal artifact reduction in CT images. IEEE Journal of Biomedical and Health Informatics, 28(11), 6712–6724. doi:10.1109/JBHI.2024.3439729
Castets, S., Thomas-Teinturier, C., Villanueva, C., Amsellem, J., Barat, P., Brun, G., Quoc, E. B., Carel, J. C., De Filippo, G. P., Kipnis, C., et al. (2024). Diagnosis and management of congenital hypopituitarism in children. Archives de Pédiatrie, 31(3), 165–171. doi:10.1016/j.arcped.2024.01.003
Chen, C., Mat Isa, N. A., & Liu, X. (2025). A review of convolutional neural network-based methods for medical image classification. Computers in Biology and Medicine, 185, 109507. doi:10.1016/j.compbiomed.2024.109507
Chmiel, W., Kwiecień, J., & Motyka, K. (2023). Saliency map and deep learning in binary classification of brain tumours. Sensors (Basel), 23(9), 4543. doi:10.3390/s23094543
Deger, J., Goethe, E. A., LoPresti, M. A., & Lam, S. (2021). Intraventricular hemorrhage in premature infants: a historical review. World Neurosurgery, 153, 21–25. doi:10.1016/j.wneu.2021.06.043
Dubois, J., Alison, M., Counsell, S. J., Hertz-Pannier, L., Hüppi, P. S., & Benders, M. J. N. L. (2021). MRI of the neonatal brain: a review of methodological challenges and neuroscientific advances. Journal of Magnetic Resonance Imaging, 53(5), 1318–1343. doi:10.1002/jmri.27192
Elerian, A. E., Rodriguez-Sanz, D., Elsherif, A. A., Dorgham, H. A., Al-Hamaky, D. M. A., Fakharany, M. S. E., & Ewidea, M. (2024). A comparative analysis of high-intensity laser therapy vs. shock wave therapy in diabetic frozen shoulder management. Journal of Medical Sciences Interdisciplinary Research, 4(2), 41–46. doi:10.51847/HA5MUZmTk4
Fatima, N., Khan, U., Han, X., Zannin, E., Rigotti, C., Cattaneo, F., Dognini, G., Ventura, M. L., & Demi, L. (2024). Deep learning approaches for automated classification of neonatal lung ultrasound with assessment of human-to-AI interrater agreement. Computers in Biology and Medicine, 183, 109315. doi:10.1016/j.compbiomed.2024.109315
Feferman, B., Fernandez, V., Tavernier, E., Fuseau, J., Sembély-Taveau, C., Boennec, R., Cremades, A., Maakaroun-Vermesse, Z., Follet, C., Mitanchez, D., et al. (2025). Diagnostic imaging of Bacillus cereus brain infection in newborns. Pediatric Radiology, 55(4), 835–845. doi:10.1007/s00247-025-06169-7
Fenton, L. Z. (2022). Imaging of congenital malformations of the brain. Clinics in Perinatology, 49(3), 587–601. doi:10.1016/j.clp.2022.05.002
Fortin, O., Christoffel, K., Shoaib, A. B., Venkatesan, C., Cilli, K., Schroeder, J. W., Alves, C., Ganetzky, R. D., & Fraser, J. L. (2024). Fetal brain MRI abnormalities in pyruvate dehydrogenase complex deficiency. Neurology, 103(4), e209728. doi:10.1212/WNL.0000000000209728
G, V., Rani, V. V., Ponnada, S., & S, J. (2025). A hybrid EfficientNet-DbneAlexnet for brain tumor detection using MRI images. Computational Biology and Chemistry, 115, 108279. doi:10.1016/j.compbiolchem.2024.108279
Ganea, M., Horvath, T., Nagy, C., Morna, A. A., Pasc, P., Szilagyi, A., Szilagyi, G., Sarac, I., & Cote, A. (2024). Rapid method for microencapsulation of Magnolia officinalis oil and its medical applications. Special Journal of Pharmacognosy, Phytochemistry & Biotechnology, 4, 29–38. doi:10.51847/UllqQHbfeC
Gombolay, G. Y., Gopalan, N., Bernasconi, A., Nabbout, R., Megerian, J. T., Siegel, B., Hallman-Cooper, J., Bhalla, S., & Gombolay, M. C. (2023). Review of machine learning and artificial intelligence (ML/AI) for the pediatric neurologist. Pediatric Neurology, 141, 42–51. doi:10.1016/j.pediatrneurol.2023.01.004
Gu, Z., Dogra, S., Siriruchatanon, M., Kneifati-Hayek, J., & Kang, S. K. (2025). Radiology workflow assistance with artificial intelligence: Establishing the link to outcomes. Journal of the American College of Radiology. Advance online publication. doi:10.1016/j.jacr.2025.10.018
Guarnera, A., Moltoni, G., Dellepiane, F., Lucignani, G., Rossi-Espagnet, M. C., Campi, F., Auriti, C., & Longo, D. (2024). Bacterial meningoencephalitis in newborns. Biomedicines, 12(11), 2490. doi:10.3390/biomedicines12112490
Hamid, A. A., Abdelkareemm, M. A., Al Nabulsi, Y. A. B., & Hamad, A. H. (2022). The role of collaborative climate in fostering knowledge sharing: evidence from Sudanese insurance companies. Annals of Organizational Culture, Leadership & External Engagement Journal, 3, 65–72. doi:10.51847/CJyBigOMA1
Holste, K. G., Xia, F., Ye, F., Keep, R. F., & Xi, G. (2022). Mechanisms of neuroinflammation in hydrocephalus after intraventricular hemorrhage: a review. Fluids and Barriers of the CNS, 19(1), 28. doi:10.1186/s12987-022-00324-0
Hwang, M., Tierradentro-García, L. O., Hussaini, S. H., Cajigas-Loyola, S. C., Kaplan, S. L., Otero, H. J., & Bellah, R. D. (2022). Ultrasound imaging of preterm brain injury: fundamentals and updates. Pediatric Radiology, 52(4), 817–836. doi:10.1007/s00247-021-05191-9
Karthikeyan, V., Muthupriya, P., Gopikrishna, M., & Sivakumar, K. (2024). Effects of electromagnetic radiation and radio frequency on freshwater calanoid and cyclopoid copepods. World Journal of Environmental Biosciences, 13(2), 1–5. doi:10.51847/YYlqFBgHxk
Khandia, R., Ali Khan, A., Alexiou, A., Povetkin, S. N., & Verevkina, M. N. (2022). Codon usage analysis of pro-apoptotic Bim gene isoforms. Journal of Alzheimer’s Disease, 86(4), 1711–1725. doi:10.3233/JAD-215691
Khandia, R., Pandey, M. K., Zaki, M. E. A., Al-Hussain, S. A., Baklanov, I., & Gurjar, P. (2023). Application of codon usage and context analysis in genes up- or down-regulated in neurodegeneration and cancer to combat comorbidities. Frontiers in Molecular Neuroscience, 16, 1200523. doi:10.3389/fnmol.2023.1200523
Khazaal, G., Daou, M., Mahdi, S. S., Ahmed, Z., Maalouf, E., Batteni, G., Qasim, S. S., Kassis, C., Agha, D., Haddad, H., et al. (2023). Comparison of color stability in SDR flowable material and packable composite using Easy-Shade device. Turkish Journal of Dental Hygiene, 3, 15–21. doi:10.51847/HrmuqLFIpg
Kim, J. Y., Nam, Y., Kim, S., Shin, N. Y., & Kim, H. G. (2023). MRI-visible perivascular spaces in the neonatal brain. Radiology, 307(2), e221314. doi:10.1148/radiol.221314
Kitama, T., Nishiyama, T., Hosoya, M., Shimanuki, M. N., Ueno, M., You, F., Ozawa, H., & Oishi, N. (2022). Exploring noise-induced hearing loss: a comprehensive systematic review. Interdisciplinary Research in Medical Sciences Special, 2(2), 1–10. doi:10.51847/p7jSxCe2qx
Lagercrantz, H. (2025). The awakening of the newborn human infant and the emergence of consciousness. Acta Paediatrica, 114(5), 823–828. doi:10.1111/apa.70031
Largent, A., De Asis-Cruz, J., Kapse, K., Barnett, S. D., Murnick, J., Basu, S., Andersen, N., Norman, S., Andescavage, N., & Limperopoulos, C. (2022). Automatic brain segmentation in preterm infants with post-hemorrhagic hydrocephalus using 3D Bayesian U-Net. Human Brain Mapping, 43(6), 1895–1916. doi:10.1002/hbm.25762
Lew, C. O., Calabrese, E., Chen, J. V., Tang, F., Chaudhari, G., Lee, A., Faro, J., Juul, S., Mathur, A., McKinstry, R. C., et al. (2024). Artificial intelligence outcome prediction in neonates with encephalopathy (AI-OPiNE). Radiology: Artificial Intelligence, 6(5), e240076. doi:10.1148/ryai.240076
Li, Y., Wisnowski, J. L., Chalak, L., Mathur, A. M., McKinstry, R. C., Licona, G., Mayock, D. E., Chang, T., Van Meurs, K. P., Wu, T. W., et al. (2022). Mild hypoxic-ischemic encephalopathy (HIE): Timing and pattern of MRI brain injury. Pediatric Research, 92(6), 1731–1736. doi:10.1038/s41390-022-02026-7
Lotlikar, V. S., Satpute, N., & Gupta, A. (2022). Brain tumor detection using machine learning and deep learning: a review. Current Medical Imaging, 18(6), 604–622. doi:10.2174/1573405617666210923144739
Mancha, G. T., Kadakia, S., Muñoz, L., & Seske, L. M. (2023). Ten-year review of neonatal neurosurgical outcomes and cost analysis. Surgical Neurology International, 14, 203. doi:10.25259/SNI_59_2023
Morrison, S. L., Dukhovny, D., Chan, R. V. P., Chiang, M. F., & Campbell, J. P. (2022). Cost-effectiveness of artificial intelligence-based retinopathy of prematurity screening. JAMA Ophthalmology, 140(4), 401–409. doi:10.1001/jamaophthalmol.2022.0223
Moshfeghi, D. M. (2024). Artificial intelligence poised to improve retinopathy of prematurity screening. Ophthalmology Retina, 8(1), 1–2. doi:10.1016/j.oret.2023.09.023
Perri, A., Sbordone, A., Patti, M. L., Nobile, S., Tirone, C., Giordano, L., Tana, M., D’Andrea, V., Priolo, F., Serrao, F., et al. (2023). The future of neonatal lung ultrasound: Validation of an artificial intelligence model for interpreting lung scans. A multicentre prospective diagnostic study. Pediatric Pulmonology, 58(9), 2610–2618. doi:10.1002/ppul.26563
Pindrik, J., Schulz, L., & Drapeau, A. (2022). Diagnosis and surgical management of neonatal hydrocephalus. Seminars in Pediatric Neurology, 42, 100969. doi:10.1016/j.spen.2022.100969
Rajamani, K. T., Rani, P., Siebert, H., ElagiriRamalingam, R., & Heinrich, M. P. (2023). Attention-augmented U-Net (AA-U-Net) for semantic segmentation. Signal, Image and Video Processing, 17(4), 981–989. doi:10.1007/s11760-022-02302-3
Ramanathan, A., Athikarisamy, S. E., & Lam, G. C. (2023). Artificial intelligence for the diagnosis of retinopathy of prematurity: a systematic review of current algorithms. Eye (London), 37(12), 2518–2526. doi:10.1038/s41433-022-02366-y
Rener-Primec, Z., Neubauer, D., & Osredkar, D. (2022). Dysmature patterns of newborn EEG recordings: biological markers of transitory brain dysfunction or brain injury. European Journal of Pediatric Neurology, 38, 20–24. doi:10.1016/j.ejpn.2022.03.008
Rogers, H. J., Vermaire, J. H., Gilchrist, F., & Schuller, A. A. (2023). Studying the relationship between dental caries index and quality of life related to oral health. Turkish Journal of Dental Hygiene, 3, 1–8. doi:10.51847/nKdDFEMTud
Russ, J. B., Simmons, R., & Glass, H. C. (2021). Neonatal encephalopathy: beyond hypoxic-ischemic encephalopathy. Neoreviews, 22(3), e148–e162. doi:10.1542/neo.22-3-e148
Scarpa, E. C., Lyra, J. C., Lourenção, P. L. T. A., Hachem, A. S., Silva, G. H. S. D., Giacóia, G. R. F., Ortolan, E. V. P., Silva, C. P., Silveira, G. L. D., Bentlin, M. R., et al. (2025). Analysis of agreement between specialists for the evaluation of radiological findings of necrotizing enterocolitis. Journal of Pediatrics (Rio J), 101(1), 103–109. doi:10.1016/j.jped.2024.07.008
Sewankambo, P. R. (2024). Investigating clinical ethics consultation in Uganda: a case study at the Uganda Cancer Institute. Asian Journal of Ethics in Health and Medicine, 4, 28–43. doi:10.51847/ULP3gIQWcE
Shaukat, Z., Farooq, Q. U. A., Tu, S., Xiao, C., & Ali, S. (2022). A state-of-the-art technique to perform cloud-based semantic segmentation using deep learning 3D U-Net architecture. BMC Bioinformatics, 23(1), 251. doi:10.1186/s12859-022-04794-9
Shih, C. S., & Chiu, H. W. (2025). Automatic multi-stage classification model for fetal ultrasound images based on EfficientNet. Studies in Health Technology and Informatics, 329, 1882–1883. doi:10.3233/SHTI251262
Siddique, N., Paheding, S., Reyes Angulo, A. A., Alom, M. Z., & Devabhaktuni, V. K. (2022). Fractal, recurrent, and dense U-Net architectures with EfficientNet encoder for medical image segmentation. Journal of Medical Imaging (Bellingham), 9(6), 064004. doi:10.1117/1.JMI.9.6.064004
Sullivan, B. A., Beam, K., Vesoulis, Z. A., Aziz, K. B., Husain, A. N., Knake, L. A., Moreira, A. G., Hooven, T. A., Weiss, E. M., Carr, N. R., et al. (2024). Transforming neonatal care with artificial intelligence: Challenges, ethical considerations, and opportunities. Journal of Perinatology, 44(1), 1–11. doi:10.1038/s41372-023-01848-5
Vahedifard, F., Ai, H. A., Supanich, M. P., Marathu, K. K., Liu, X., Kocak, M., Ansari, S. M., Akyuz, M., Adepoju, J. O., Adler, S., et al. (2023). Automatic ventriculomegaly detection in fetal brain MRI: A step-by-step deep learning model for novel 2D-3D linear measurements. Diagnostics (Basel), 13(14), 2355. doi:10.3390/diagnostics13142355
Wirth, W., Eckstein, F., Kemnitz, J., Baumgartner, C. F., Konukoglu, E., Fuerst, D., & Chaudhari, A. S. (2021). Accuracy and longitudinal reproducibility of quantitative femorotibial cartilage measures derived from automated U-Net-based segmentation of two different MRI contrasts: Data from the Osteoarthritis Initiative healthy reference cohort. MAGMA, 34(3), 337–354. doi:10.1007/s10334-020-00889-7
Wisnowski, J. L., Wintermark, P., Bonifacio, S. L., Smyser, C. D., Barkovich, A. J., Edwards, A. D., de Vries, L. S., Inder, T. E., & Chau, V.; Newborn Brain Society Guidelines and Publications Committee. (2021). Neuroimaging in the term newborn with neonatal encephalopathy. Seminars in Fetal & Neonatal Medicine, 26(5), 101304. doi:10.1016/j.siny.2021.101304
Wu, Y. W., Monsell, S. E., Glass, H. C., Wisnowski, J. L., Mathur, A. M., McKinstry, R. C., Bluml, S., Gonzalez, F. F., Comstock, B. A., Heagerty, P. J., et al. (2023). How well does neonatal neuroimaging correlate with neurodevelopmental outcomes in infants with hypoxic-ischemic encephalopathy? Pediatric Research, 94(3), 1018–1025. doi:10.1038/s41390-023-02510-8
Xie, B., Liu, Y., Li, X., Yang, P., & He, W. (2023). Enhancing the dissolution rate of dolutegravir sodium using nanosuspension technology and a 32 factorial design. Pharmaceutical Science & Drug Design, 3, 12–19. doi:10.51847/2uCOYf3jPn
Xie, K., Gao, L., Zhang, H., Zhang, S., Xi, Q., Zhang, F., Sun, J., Lin, T., Sui, J., & Ni, X. (2024). GAN-based metal artifacts region inpainting in brain MRI imaging with reflective registration. Medical Physics, 51(3), 2066–2080. doi:10.1002/mp.16724
Xu, L., Yang, J., Zhang, Y., Liu, X., Liu, Z., Sun, F., Ma, Y., Wang, L., & Xing, F. (2024). Bioinformatics analysis of gene modules and key genes for the early diagnosis of gastric cancer. Archives of International Journal of Cancer and Allied Sciences, 4(1), 24–36. doi:10.51847/2gUBplgIMV
Yousef, R., Khan, S., Gupta, G., Siddiqui, T., Albahlal, B. M., Alajlan, S. A., & Haq, M. A. (2023). U-Net-based models towards optimal MR brain image segmentation. Diagnostics (Basel), 13(9), 1624. doi:10.3390/diagnostics13091624
Zhang, Q., & Zhou, X. (2023). Analysis of cranial ultrasound images for newborns. Frontiers in Neurology, 13, 1090275. doi:10.3389/fneur.2022.1090275
Zhou, J., & Dewey, R. S. (2024). The association between achievement motivation and hardiness. Journal of Advanced Pharmaceutical Education and Research, 14(2), 50–57. doi:10.51847/XsYAMPlBZc
