FUNCTIONAL ENDOSCOPIC SINUS SURGERY USING ARTIFICIAL INTELLIGENCE IN CHILDREN WITH CHRONIC POLYPOUS RHINOSINUSITIS

ABSTRACT

The study included 64 pediatric patients with chronic polypous rhinosinusitis who underwent AI-assisted preoperative planning based on automatic segmentation of anatomical structures. Neural network algorithms employing a U-Net architecture improved the accuracy of anatomical structure identification and reduced surgical preparation time. The findings indicate that the integration of artificial intelligence technologies in otorhinolaryngology enhances surgical planning precision and holds significant potential for improving the safety and effectiveness of functional endoscopic sinus surgery in children with polypous rhinosinusitis.

АННОТАЦИЯ

В исследование было включено 64 пациента детского возраста с хроническим полипозным риносинуситом, которым проводилось предоперационное планирование с использованием технологий искусственного интеллекта на основе автоматической сегментации анатомических структур. Алгоритмы нейронных сетей с архитектурой U-Net позволили повысить точность идентификации анатомических структур и сократить время подготовки к операции. Полученные результаты свидетельствуют о том, что интеграция технологий искусственного интеллекта в оториноларингологию улучшает точность хирургического планирования и обладает значительным потенциалом для повышения безопасности и эффективности функциональной эндоскопической синус-хирургии у детей с полипозным риносинуситом.

Keywords: artificial intelligence, pediatric rhinology, functional endoscopic sinus surgery, nasal polyposis.

Ключевые слова: искусственный интеллект, детская ринология, функциональная эндоскопическая синус-хирургия, полипоз носа.

Introduction

Chronic rhinosinusitis with nasal polyps (CRSwNP) in children represents a clinically significant and therapeutically challenging condition in pediatric otorhinolaryngology. The disease is characterized by persistent inflammation of the sinonasal mucosa, marked anatomical variability, a high rate of recurrence, and an increased risk of intraoperative complications. In contrast to adults, pediatric CRSwNP is frequently associated with systemic conditions such as allergic diseases, cystic fibrosis, and immunological disorders, reflecting its multifactorial pathogenesis and complex clinical course [3, p. 13].

Despite substantial advances in pharmacological management, including prolonged courses of topical corticosteroids and the introduction of biological agents targeting type 2 inflammation, a considerable proportion of pediatric patients remain refractory to conservative treatment and require surgical intervention. Functional endoscopic sinus surgery (FESS) is currently regarded as the gold standard surgical approach in such cases, aiming to restore physiological ventilation and drainage of the paranasal sinuses while preserving the integrity of the mucosal lining.

However, performing FESS in children presents specific anatomical and technical challenges. These include incomplete pneumatization of the paranasal sinuses, thin and fragile bony structures, variability of anatomical landmarks, and the close proximity of critical structures such as the orbit and the skull base [4, p. 14]. These factors significantly increase the potential risk of surgical complications and demand meticulous preoperative planning.

Traditionally, surgical planning relies on the subjective interpretation of computed tomography (CT) images by the surgeon, which may lead to variability in risk assessment and intraoperative decision-making. In recent years, the rapid development of artificial intelligence (AI) technologies, particularly in medical image analysis, has opened new opportunities for objective assessment of anatomical structures, automated segmentation, identification of risk zones, and optimization of surgical access strategies.

The integration of AI into preoperative planning for FESS may enhance anatomical visualization, improve surgical precision, reduce complication rates, and ultimately contribute to better postoperative outcomes in pediatric patients with CRSwNP. Nevertheless, the practical applicability and effectiveness of AI-assisted planning in pediatric endoscopic sinus surgery remain insufficiently investigated.

Aim of the study: To evaluate the potential of artificial intelligence applications for preoperative CT analysis and surgical access planning in functional endoscopic sinus surgery in children with chronic rhinosinusitis with nasal polyps.

Materials and methods

The study included 64 children diagnosed with chronic polypous rhinosinusitis who were hospitalized in the ENT department of the Tashkent State Medical University Children's Clinic during 2022-2025. The patients' ages ranged from 9 to 17 years, with a mean age of 11.8 ± 2.9 years. Among those examined were 38 boys (59.4%) and 26 girls (40.6%). All patients were divided into two clinical groups based on the surgical approach used:

Group I (control) - 32 children who underwent standard functional endoscopic sinus surgery (FESS) according to the generally accepted method based on traditional analysis of computed tomography data.

Group II (main) - 32 children who underwent surgical treatment using artificial intelligence for preoperative FESS planning, including automatic segmentation of nasal cavity anatomical structures, three-dimensional reconstruction of paranasal sinuses, and determination of optimal endoscopic access.

Inclusion criteria for the study: confirmed diagnosis of chronic rhinosinusitis with nasal polyps, age 9 to 17 years, lack of effect from conservative therapy for at least 12 weeks, availability of paranasal sinus computed tomography data, and informed consent from parents or legal representatives.

Exclusion criteria: acute infectious diseases during the examination period, congenital facial skeleton defects, malignant neoplasms, previous extensive operations on the paranasal sinuses, and severe somatic diseases in the decompensation stage.

All patients underwent computed tomography of the paranasal sinuses with slice thickness of 0.5-0.6 mm. High-resolution computed tomograms of the paranasal sinuses were analyzed using deep machine learning algorithms. Automatic segmentation of anatomical structures, three-dimensional reconstruction, and mapping of surgical risk zones were performed. Individual surgical trajectories were developed based on quantitative assessment of skull base structure parameters, lamina papyracea, location of ethmoidal arteries, and extent of the inflammatory process.

In the control group, surgical intervention was performed using the standard FESS method, focusing on anatomical landmarks and visual analysis of CT images. In the main group, an AI-assisted preoperative planning system was additionally used, which included: automatic segmentation of the paranasal sinuses, identification of individual anatomical variants, mapping of areas with increased surgical risk, modeling of the optimal endoscopic access pathway, and integration with the navigation system during the operation. The extent of surgical intervention was determined individually, considering the prevalence of the polypoid process and the patient's anatomical features.

Statistical processing of the obtained data was carried out using SPSS Statistics 26.0 software.

Results and discussions

To minimize systematic error and ensure correct statistical comparison, patients were divided into control and main groups with maximally similar numerical distributions. This approach ensured the comparability of groups in terms of sample size and increased the reliability of assessing differences identified during the analysis of treatment results.

Analysis of initial clinical and demographic indicators showed that the control and main groups were comparable in age, gender, prevalence of bilateral polypoid rhinosinusitis, and presence of allergic rhinitis. For all studied parameters, p-values exceeded the established threshold of statistical significance (p > 0.005), indicating the absence of statistically significant differences between the groups at the initial stage of the study. Thus, the homogeneity of the groups allows for a correct assessment of treatment effectiveness and reliable interpretation of the obtained results (Table 1).

Table 1.

Main parameters in observed patients before treatment

The use of AI-assisted planning based on automatic segmentation of anatomical structures significantly optimizes the preoperative preparation process. The application of neural network algorithms (particularly the U-Net architecture) has provided high accuracy in contouring the paranasal sinuses and critically important risk zones, which minimizes the probability of iatrogenic injuries (Table 2).

Table 2.

Comparative characteristics of segmentation accuracy in patients with chronic polypoid rhinosinusitis

The average Dice similarity coefficient for the segmentation of the paranasal sinuses in the main group was 0.91±0.03, for the osteomeatal complex - 0.89±0.04, for the lamina papyracea - 0.92±0.02, and for the skull base and cribriform plate - 0.90±0.03, which significantly exceeded the indicators of manual segmentation in the control group (0.82±0.05; p<0.01).

Automatic identification of inflammatory and polypoid tissue enabled a more accurate determination of the pathological substrate volume. The average volume of polypoid tissue, calculated based on three-dimensional reconstruction, was 6.8±1.2 cm3, with the discrepancy between intraoperative and preoperative assessment in the main group not exceeding 7.4%, while in the control group this indicator reached 18.9% (p<0.01).

The use of individual 3D models of the paranasal sinuses contributed to a decrease in the frequency of intraoperative complications. Damage to the lamina papyracea was recorded in 3 patients (9.4%) of the control group and was not observed in the main group (0%; p<0.05). The frequency of intraoperative bleeding decreased from 21.9% (7 cases) in the control group to 6.3% (2 cases) in the main group (p<0.05).

Furthermore, the use of AI-assisted planning made it possible to reduce the average duration of surgical intervention from 78.5±9.6 min in the control group to 61.2±8.4 min in the main group (p<0.01), as well as to decrease the frequency of postoperative synechiae from 31.3% to 9.4% (p<0.05).

Thus, the automatic segmentation of anatomical structures and the construction of individual three-dimensional models of paranasal sinuses in patients with chronic polypoid rhinosinusitis ensured increased accuracy of preoperative planning, reduced intra- and postoperative complications, and improved clinical outcomes of surgical treatment.

The diagram below clearly illustrates the impact of AI-assisted planning on the results of functional endoscopic sinus surgery (FESS) in children with chronic polypoid rhinosinusitis. The frequency of intraoperative complications, particularly bleeding, decreased from 21.9% in the control group to 6.3% in the main group, indicating a significant increase in the safety of surgical intervention when using AI support in preoperative planning. Damage to the lamina papyracea was observed only in the control group (9.4%) and was absent in the main group, demonstrating the effectiveness of AI in preventing injury to thin bone structures (Fig. 1).

Figure 1. Comparative characteristics of postoperative complications in the main and control groups

Additionally, the formation of postoperative synechiae decreased from 31.3% in the control group to 9.4% in the main group, which indicates better preservation of the mucous membrane and more careful execution of FESS when using AI. Recurrence of polyposis after 12 months was observed in 40.6% of patients in the control group and only in 15.4% of patients in the main group, confirming the long-term effectiveness of AI-assisted planning in reducing the frequency of recurrent polyposis.

Thus, the use of AI in preoperative planning of FESS allows for a significant reduction in the frequency of intraoperative complications, postoperative synechiae, and relapses, ensuring safer and more effective surgical intervention in children.

Conclusions

The use of artificial intelligence technologies in functional endoscopic surgery of the paranasal sinuses in children with polypous rhinosinusitis is a promising and scientifically grounded approach.

The use of neural network algorithms for automatic segmentation of anatomical structures contributes to accurate navigation during surgery, reduces the risk of damage to critical tissues, and shortens the operating time, which is especially important during surgical interventions in children.

AI-assisted surgical planning provides a personalized, organ-preserving, and safe approach that allows for improved immediate and long-term treatment outcomes.

References:

1. Alimova DD, Sagidullaeva DB. The clinical characteristics of patients and treatment of chronic polypous rhinosinusitis. Cent Asian J Med. 2025;2:32-41.
2. Amonov S, Alimova D, Nurmukhamedova F, Umarov K, Salomov K. Treatment optimization of chronic odontogenic maxillary sinusitis. Eur J Mol Clin Med. 2020;7(2):2419-2422.
3. Demirbas D, Cingi C, Cakli H, Kaya E. Use of rhinomanometry in common rhinologic disorders. Expert Rev Med Devices. 2011.
4. Fedina IN, Pankova VB, Serebryakov PV. Pathology of upper respiratory tract: professional risk formation. Russ Rhinol. 2018;26(4):35-39.
5. Fokkens WJ, Rosenfeld RM, et al. European Position Paper on Rhinosinusitis and Nasal Polyps. Rhinology. 2013;45(20):1–139.
6. Khalilov NM, Madaminova MA. Feasibility of simultaneous functional and aesthetic interventions in rhinoplasty and nasal cavity surgeries. Vestn KRSU. 2020;20(5):95-99.
7. Madaminova MA, Nuraliev MA, Nasyrov VA, Izaeva TA, et al. Lektsii po otorinolaringologii. Kurs lektsiy. 2022; 41.25 p.l.
8. Mamatova Sh, Karabaev Kh, Namakhanov A. Ultrasound investigation of rhinosinusitis in early childhood. J Biomed Pract. 2023;1(2):63–69. https://doi.org/10.26739/2181-9300-2021-2-10
9. Nasyrov VA, Izaeva TA, Nasyrov MV, Solodchenko NV. Verkhnchelyustnoy sinusit. Monograph. Bishkek; 2023. 14.75 p.l.
10. Nurmukhamedova FB. Clinical aspects of allergic rhinitis in children with parasitic invasion. Saratov Sci Med J. 2018;1.
11. Rajabov AH, Amonov ShE, Inoyatova FI, Amonov AS. Condition of ENT organs in children with chronic hepatitis B. Vrach-aspirant. 2009;31(4):323-327.
12. Rajabov AH, Amonov ShE. Diagnosis and treatment of nasal bleeding in children with chronic hepatitis B. Otorhinolaryngology. 2020;4.
13. Rajabov AH, Inoyatova FI, Amonov ShE. Features of clinical course of chronic tonsillitis in children with chronic hepatitis. Russ Otorhinolaryngol. 2012;6:132-136.
14. Rajabov AH, Toshtemirova MM. Features of chronic rhinosinusitis in children. Eurasian J Otorhinolaryngol Head Neck Surg. 2025;4:42-46.