Three‐dimensional image analysis for staging chronic rhinosinusitis

Abstract

Background Traditional methods of staging chronic rhinosinusitis (CRS) through imaging do not differentiate between degrees of partial mucosal sinus inflammation, thus limiting their utility as imaging biomarkers. We hypothesized that software-aided, quantitative measurement of sinus inflammation would generate a metric of disease burden that would correlate with clinical parameters in patients with suspected sinus disease. Methods Adults with rhinologic complaints undergoing CT imaging were recruited at an urban, academic, tertiary care center (n=45 with Lund-Mackay [LM] scores ≥ 4). 3D volumetric image analysis was performed using a semi-automated method to obtain a “Chicago-modified Lund-Mackay” (Chicago MLM) score, which provides a continuous scale to quantify extent of opacification. Linear regression was used to test the association of the Chicago MLM score with concurrent symptoms (total nasal symptom scores [TNSS]) and disease-specific quality of life (Sinonasal Outcome Test-22 [SNOT22]). Results Chicago MLM scores were significantly associated with both symptoms (p=0.037) and disease-specific quality of life (p=0.007). Inflammation in the ethmoid and sphenoid sinuses appeared to influence these associations. These findings were even more robust when analysis was limited to patients with more severe disease (LM>6). Conclusions The quantitative measurement of sinus inflammation by computer-aided 3D analysis correlates modestly with both symptoms and disease-specific quality of life. Posterior sinuses appear to have the greatest impact on these findings, potentially providing an anatomic target for clinicians to base therapy. The Chicago MLM score is a promising imaging biomarker for clinical and research use.

Volumetric assessment of paranasal sinus opacification on computed tomography can be automated using a convolutional neural network

“…Similarly, Likness et al compared multiple visual CT scoring systems with volumetric scoring based on manual CT segmentation and found that volumetric analysis was more sensitive to therapeutic effect 2 . Separately, Garneau et al 31 and Lim et al 3 each showed correlation between volumetric opacification scores and symptoms assessed using the Total Nasal Symptoms score and the 22‐item Sino‐Nasal Outcome Test. Although the results are promising, all of these efforts relied on manual or semiautomatic segmentation of the sinus cavities, which can take anywhere from 20 minutes 32 to several hours 33 to accomplish.…”

“…There is a recognized unmet need for precise, objective evaluation of disease severity, and treatment response in chronic rhinosinusitis (CRS) 1‐3 . Computed tomography (CT) is the preferred modality for noninvasive imaging of the paranasal sinuses and plays a key role in evaluating CRS.…”

“…Visual scoring is also time‐consuming, subjective, and lacks sensitivity to change. Computer approaches for quantification of sinus cavity contents on CT have shown promise, 1‐3 but existing methods rely on manual or semiautomatic segmentation of the paranasal sinus cavities, which is also time‐consuming and not practical for routine clinical use. A fully automated system for sinus segmentation on CT could facilitate efficient and truly objective quantification of sinonasal inflammation, which may have clinical utility as a screening tool or could enable identification of new phenotypes or markers of severity in upper and lower airway diseases.…”

Automated classification of osteomeatal complex inflammation on computed tomography using convolutional neural networks

“…As a result, the literature regarding correlation of these CT scoring systems and patient-reported outcomes has been mixed 8 . Interestingly, more recent studies using semi-automated volumetric analysis of CT scans have shown a higher degree of association with symptoms 9,10 , suggesting that more comprehensive methods of CT image analysis may be important for linking symptoms with objective measures of inflammation. These methods, however, are time-consuming and may not be practical for routine clinical use.…”

“…Briefly, a CNN is a biologically-inspired computer algorithm that uses thousands of simple units, modeled after neurons, that take an input from one layer of units, perform a mathematical transformation, and feed the output to the next set of neurons. Similar to their biological counterparts, while each individual neuron is simple to understand, structured networks of them have demonstrated complicated, emergent properties that are useful for pattern-recognition tasks 9,10 . These algorithms have been used to great effect in multiple applications 11 , most notably the ImageNet challenge, an annual international computer vision competition that asks computer scientists to develop algorithms to classify a given set of images into over 1,000 categories.…”