Semi-Supervised Deep Learning Semantic Segmentation for 3D Volumetric Computed Tomographic Scoring of Chronic Rhinosinusitis: Clinical Correlations and Comparison with Lund-Mackay Scoring

Abstract

Background: The traditional Lund-Mackay score (TLMs) is unable to subgrade the volume of inflammatory disease. We aimed to propose an effective modification and calculated the volume-based modified LM score (VMLMs), which should correlate more strongly with clinical symptoms than the TLMs. Methods: Semi-supervised learning with pseudo-labels used for self-training was adopted to train our convolutional neural networks, with the algorithm including a combination of MobileNet, SENet, and ResNet. A total of 175 CT sets, with 50 participants that would undergo sinus surgery, were recruited. The Sinonasal Outcomes Test-22 (SNOT-22) was used to assess disease-specific symptoms before and after surgery. A 3D-projected view was created and VMLMs were calculated for further comparison. Results: Our methods showed a significant improvement both in sinus classification and segmentation as compared to state-of-the-art networks, with an average Dice coefficient of 91.57%, an MioU of 89.43%, and a pixel accuracy of 99.75%. The sinus volume exhibited sex dimorphism. There was a significant positive correlation between volume and height, but a trend toward a negative correlation between maxillary sinus and age. Subjects who underwent surgery had significantly greater TLMs (14.9 vs. 7.38) and VMLMs (11.65 vs. 4.34) than those who did not. ROC-AUC analyses showed that the VMLMs had excellent discrimination at classifying a high probability of postoperative improvement with SNOT-22 reduction. Conclusions: Our method is suitable for obtaining detailed information, excellent sinus boundary prediction, and differentiating the target from its surrounding structure. These findings demonstrate the promise of CT-based volumetric analysis of sinus mucosal inflammation.

A Scoping Review of Artificial Intelligence Research in Rhinology

“…Of the 79 articles identified, 3 were published in 2017, 17–19 5 were published in 2018, 20–24 9 were published in 2019, 25–33 15 were published in 2020, 34–48 31 were published in 2021, 49–79 and 16 were published in 2022 80…”

“…Articles were published in journals of otolaryngology (n = 19), 17,18,23,26,28,30,36,38,39,42,46,52,57,65,72,77,86,87,94 radiology (n = 21), 19,22,25,29,31,33,34,45,48,51,54,55,58,67,73,74,81,83–85,95 medical sciences (n = 19), 32,41,44,47,53,61,63,66,68–70,76,78–80,89–91,93 or other areas of medicine (n = 20). 20,21,24,27,35,37,…”

“…Radiology Diagnostics There were 42 articles that used an application of AI for radiology/imaging diagnostics in rhinology. 19,21,22,25,26,29,31,38–43,45,46,48–51,54,55,58,59,62–64,67,69,75,78,80–86,89,91–93,95 Imaging tools used include CT (n = 25), 21,22,25,26,38–43,45,46,55,58,59,63,69,75,81–83,89,91,92,…”

“…The remaining 27 articles aimed to identify or differentiate between various pathologies 19,26,29,31,39,40,42,46,48,49,54,55,58,62–64,67,69,75,80–86,89 including concha bullosa (n = 1), 42 CRS (n = 3), 39,46,83 fungal ball (n = 1), 82 inverted papilloma (n = 3), 19,69,86 nasal cancers (n = 4), 19,49,63,86 maxillary defects (n = 1), 75 maxillary sinus lesions (n = 3), 55,58,81 maxillary sinusitis (n = 6), 29,31,54,62,64,85 nasal fractures (n = 1), 89 nasopharyngeal cancers (n = 4) 40,48,67,80 or other pathologies (n = 1), 84 and osteomeatal complex inflammation (n = 1). 26 Diagnostic accuracy was rated as excellent (n = 12), 39,40,42,46…”

“…Of the 79 articles identified, 3 were published in 2017, [17][18][19] 5 were published in 2018, [20][21][22][23][24] 9 were published in 2019, [25][26][27][28][29][30][31][32][33] 15 were published in 2020, [34][35][36][37][38][39][40][41][42][43][44][45][46][47][48] 31 were published in 2021, and 16 were published in 2022 [80][81][82][83][84][85][86][87][88][89][90][91][92][93][94][95] until May 14th (Figure 3). Study designs of the included articles were either cross-sectional (n = 61) 1...…”