Avaliação tomográfica dos seios paranasais em população brasileira: revisão sistemática da estimativa sexual, classificação dos seios esfenoidais e estudo com inteligência artificial

Abstract

Human identification represents one of the most relevant aspects in the field of forensic sciences. Various imaging and assessment methods have been employed for sex estimation. This study aimed to: (1) summarize the scientific evidence on the accuracy of linear and volumetric measurements of the maxillary sinus (MS) and frontal sinus (FS) in multislice computed tomography (MSCT) through a systematic review with meta-analysis; (2) evaluate the relationships between types of sphenoid sinus (SS), septation, lobulation, symmetry, septal deviation, and their variations in pneumatization concerning surrounding neurovascular structures in MSCT; (3) assess the specificity, sensitivity, and accuracy of MS, FS, and SS measurements using artificial intelligence (AI). In Chapter 1, a systematic review was conducted, registered in PROSPERO, and consulted six databases and gray literature. The meta-analysis of linear and volumetric data used a random-effects model and mean differences in 18 studies (2507 MSCTs), which were deemed eligible for meta-analysis and meta-regression. The width of the FS (-4.89 mm; p < 0.00001), the height of the MS (-3.73 mm; p <0.00001), and the volumes of the FS (-3,01 cm3; p < 0,00001) and MS (-1,04 cm3; p < 0,00001) exhibited the largest mean differences between sexes. Both sinuses proved to be dimorphic structures. In Chapter 2, an observational, longitudinal, retrospective, and analytical study was conducted to investigate the relationships between SS types and various anatomical variations. Findings revealed an incidence of pneumatization variations ranging from 0.3% (conchal) to 60% (post-sellar). The extension variants included 72.5% subdorsal and 92.2% lobular. The high presence of complete septa (90.9%) divides the sinus into different antra. The internal carotid artery (ICA) and the optic nerve (ON) were identified. As the SS pneumatizes, the frequency of ON protrusion, ICA prominence, and bony wall dehiscence increases. The anatomical variations of the SS are crucial for diagnosing and understanding sinus pathologies. In Chapter 3, the accuracy of sex estimation using MSCT sinus measurements in Brazilian adult individuals was evaluated through multivariate statistics and AI. A cross-sectional, retrospective study was conducted with an estimated sample of 220 MSCT scans (113 males and 107 females), aged between 18 and 49 years (43.13 ± 17.39 years), to assess the following sinus measurements: (a) volume, (b) supero-inferior, (c) latero-lateral, (d) anteroposterior, and (e) supero-anterior. Sinus evaluations were performed using a semi-automatic segmentation method with ITK-SNAP (version 4.2.0) and 3D-Slicer (version 5.2.0) software. Sensitivity, specificity, and area under the curve were calculated, along with ten different machine learning algorithms. The highest sensitivity values were observed in the three-dimensional anteroposterior (73.7%) and supero-anterior (71.1%) measurements. Among the machine learning algorithms, linear SVM and logistic regression analyses achieved the highest sex estimation accuracy. Artificial neural networks allowed for an accuracy of 84%. Therefore, AI proved to be an excellent tool in identifying predictors for sex estimation in a sample of Brazilian individuals.