Dr. HARSHAVARDHAN GHORPADE

Dr. MOREKER SUNIL RATILAL

Semi Finals

Abstract

Purpose: To study Conventional Artificial intelligence tools Vs Wolfram language analysis of papilloedema for prognosis prediction in surgical and non surgical management

Methods: 291 eyes with papilloedema due to various reasons were analysed by artificial intelligence and Wolfram language analysis. Multiple regression analysis was performed on these cases to analyse factors which are important in prognosis and correlated with clinical tools. The predictions made by the algorithms were compared with actual results in actual cases. Results: The predictive algorithm has a sensitivity of 89 % and specificity of 82 % with conventional AI and the wolfram language analysis has sensitivity of 84 % and specificity of 89 % And when combined the algorithms were able to predict the results with 90% accuracy. Conclusion: Combination of conventional AI and deep machine learning by Wolfram language can be used to predict whether surgery is needed and can be correlated with clinical tools.

Full Text

Introduction: Detection of papilloedema and the ability to determine that the optic disk is normal is valuable in the evaluation of patients with headache and other neurologic symptoms. Diagnosis of papilloedema on ophthalmoscopy determines the diagnostic strategy and treatment options. Mistakes made in detecting papilloedema usually can cause visual loss. Artificial intelligence is a useful adjunct in clinical practice. The use of artificial intelligence to detect papilloedema from fundus photographs has not been well studied.

Purpose: To study Conventional Artificial Intelligence tools Vs Wolfram language analysis of papilloedema for prognosis prediction in surgical and non surgical management.

Methods: At four tertiary centers in Mumbai and Navi Mumbai we trained, validated and tested a Wolfram language deep-learning system to classify optic disks as being normal or having papilloedema or other abnormalities from 16,783 fundus pictures taken after pupillary dilatation.15, 873 from 21 sites in 11 hospitals were used for training and validation. Performance at classifying the optic-disk appearance was evaluated by calculating the area under the receiver-operating-characteristic curve (AUC), sensitivity and specificity as compared with a reference standard of clinical diagnoses by two separate ophthalmologists.

291 eyes with due to various reasons were analysed by standard artificial intelligence and Wolfram language analysis. Multiple regression analysis was performed on these cases to analyse factors which are important in prognosis and correlated with clinical tools. The predictions made by the algorithms were compared with actual results in actual cases.

Results: In the validation set, both systems discriminated  papilloedema from normal disks with an AUC of 0.98 (95% confidence interval [CI], 0.97 to 0.99) and normal from abnormal disks with an AUC of 0.97 (95% CI, 0.96 to 0.99).The predictive algorithm has a sensitivity of 89 % and specificity of 82 % with conventional Artificial intelligence and the Wolfram language analysis has sensitivity of 84 % and specificity of 89 %.And when combined, the algorithms were able to predict the results with 90% accuracy. The factors which were found to contribute most were age, weight, myopia / hypermetropia, type of ailment, duration of ailment, visual acuity, duration of treatment.

Discussion: Many studies have discovered that direct ophthalmoscopy can be replaced by ocular fundus digital cameras that provide high-quality photographs of the optic nerve and retina, even without pharmacologic dilatation of the pupil. Most deep-learning research in ophthalmology has been for screening of retinal disorders and glaucoma. Some authors have shown that deep-learning systems could recognize right from left optic disks in the presence of optic-nerve abnormalities on fundus photographs,could discriminate disks with papilloedema from normal disks with an average accuracy of 93% (similar to the value in our study 37) and could differentiate true optic-disk swelling from pseudo-swelling with an accuracy of approximately 95%.

Wolfram language is a simple language recently developed for easier interpretation.

None have used Wolfram language as yet for papilloedema and ours is the first study to do so.

 Conclusion: Combination of conventional Artificial Intelligence and deep machine learning by Wolfram language can be used to predict whether surgery is needed and can be correlated with clinical tools.

References:-

• Bruce BB, Lamirel C, Wright DW, et al. Nonmydriatic ocular fundus photography in the emergency department. N Engl J Med 2011;364:387-389.
• Biousse V, Bruce BB, Newman NJ. Ophthalmoscopy in the 21st century: the 2017 H. Houston Merritt Lecture. Neurology 2018;90:167-175.
• Ting DSW, Peng L, Varadarajan AV, et al. Deep learning in ophthalmology: the technical and clinical considerations. Prog Retin Eye Res 2019;72:100759-100759.
• Abràmoff MD, Lavin PT, Birch M, Shah N, Folk JC. Pivotal trial of an autonomous AI-based diagnostic system for detection of diabetic retinopathy in primary care offices. NPJ Digit Med 2018;1:39-39.
• Gulshan V, Peng L, Coram M, et al. Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs. JAMA 2016;316:2402-2410.
• Liu H, Li L, Wormstone IM, et al. Development and validation of a deep learning system to detect glaucomatous optic neuropathy using fundus photographs. JAMA Ophthalmol2019;137:1353-1360.
• Liu TYA, Ting DSW, Yi PH, et al. Deep learning and transfer learning for optic disc laterality detection: Implications for machine learning in neuro-ophthalmology. J Neuroophthalmology 2019 August 22
• Akbar S, Akram MU, Sharif M, Tariq A, Yasin UU. Decision support system for detection of papilloedema through fundus retinal images. J Med Syst 2017;41:66-66.
• Ahn JM, Kim S, Ahn KS, Cho SH, Kim US. Accuracy of machine learning for differentiation between optic neuropathies and pseudo. BMC Ophthalmol2019;19:178-178.