Proptosis severity detection on Orbit CT scan with using image processing and supervised learning

Background and aims: Graves’ ophthalmopathy (GO) is a chronic autoimmune disease affectingthe retrobulbar tissues and extraocular muscles; ۴۰–۶۰% of patients may experience extraocularmuscle dysfunction. Hence, the early diagnosis of extraocular muscle abnormalities by orbital imagingmay be essential for effectively managing thyroid myopathy. It is unlikely that orbital imagingwill be performed in clinical practice unless there is a complaint of double vision. Additionally,radiologists may not always be available to interpret the results, especially in remote areas and developingcountries. Therefore, this study aims to develop a diagnostic software system to evaluatethe severity of proptosis and enlarged extraocular muscles (EEM) on nonenhanced computed tomography(NCCT) in patients with Graves’ ophthalmopathy (GO) by a deep neural network.Method: This retrospective observational study recruited ۲۰۰۰۰ participants (۲۰۰۰۰ EEM patientswith GO) whose extraocular muscle thickness of medial and lateral recti will be measuredon axial scans. The maximum thickness of the superior rectus–levator palpebrae superioris complexand inferior rectus will also be measured on the coronal scans. Moreover, on axial scansin the midglobe slice, the distance between the inter-zygotic line and the posterior and anteriorocular surfaces will be calculated to determine the position of the globe. Our cutoff values for thelength of the inter-zygomatic line and distance of the line from the posterior and anterior scleraare ۹۷mm, ۵.۹ mm, and ۲۳mm, respectively.Afterward, we began by preprocessing the images to remove noise and artifacts and then extractedrelevant features using image processing techniques. We used a supervised learning algorithmto train a model to detect proptosis severity based on the extracted features. We evaluated theperformance of our system using a separate set of validation images.Result: this article is a work in progress but in accordance with similar articles we are anticipatinga more efficient model than others and more than ۸۰% efficiency.Conclusion: Our study demonstrates that image processing and supervised learning can be usedto develop an accurate and reliable system for automated detection of proptosis severity in orbitalCT scans of patients with Graves Ophthalmopathy. The proposed system can provide an objectiveand consistent method for evaluating the severity of proptosis, which can aid in the diagnosis andtreatment of this condition. Future research may explore ways to improve the performance of oursystem further, including the incorporation of additional imaging modalities or clinical data.

کلیدواژه ها:

artificial intelligence ، Graves’ ophthalmopathy ، neural network ، proptosis