artifical intelligence in ophthalmology.pptx

artificial intelligence in ophthalmology

1. ARTIFICIAL INTELLIGENCE IN OPHTHALMOLOGY PRESENTER: DR. HARSHIKA 1

2. STAGES OF INDUSTRIAL REVOLUTION 2

3. What is Artificial intelligence ? Artificial intelligence (AI) - Technology that enables computers and machines to simulate human intelligence and problem-solving capabilities. Examples in daily life- 3 Taxi booking app Voice assistance Entertainment streaming app Image recognition through google lens Navigation and Travel

4. History of AI T erm “artificial intelligence” (AI): coined on August 31, 1955 John McCarthy , Marvin L. Minsky, Nathaniel Rochester, and Claude E. Shannon submitted “A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence.” * First scientific step towards intelligent machines: Alan Turing (1950) - famous Turing test This involves an interview with open-ended questions to determine whether the intelligence of the interviewee is human or artificial. If this distinction can no longer be made, within certain predefined margins, true machine intelligence has been accomplished. Benet D, Pellicer -Valero OJ. Artificial intelligence: the unstoppable revolution in ophthalmology. Surv Ophthalmol . 2022 Jan-Feb;67(1):252-270. doi : 10.1016/j.survophthal.2021.03.003. Epub 2021 Mar 16. PMID: 33741420 Mitchell M. Artificial intelligence: a guide for thinking humans. Penguin UK; 2019* Topol E. Deep medicine: how artificial intelligence can make healthcare human again. New York: Basic Books; 2019 Jiang F, Jiang Y, Zhi H, Dong Y, Li H, Ma S, et al. Artificial intelligence in healthcare: Past, present and future. Stroke Vasc Neurol 2017;2;2:230‑43** 4

5. 5 Tong et al; Application of machine learning in o phthalmic imaging modalities; Eye and Vision; 2020

6. Types of ai: weak Ai vs strong ai Weak/ narrow Al : Trained and focused to perform specific tasks. Most of the Al that surrounds us today are weak AI. Examples: Apple's Siri, Amazon's Alexa Strong Al/ Artificial super intelligence (ASD): Is a theoretical form of Al where a machine would have an intelligence equal to humans. Strong Al is still entirely theoretical with no practical examples in use today. Examples: Science fiction, such as HAL, the superhuman and rogue computer assistant in 2001: A Space Odyssey. 6

7. ARTIFICIAL INTELLIGENCE TECHNIQUES Tong et al; Application of machine learning in o phthalmic imaging modalities; Eye and Vision; 2020 7

8. Ai in Medicine 8

9. APPLICATIONS OF AI IN OPHTHALMOLOGY ANTERIOR SEGMENT : Keratoconus and other Anterior segment disorders Amblyopia,squint surgeries Refractive surgeries Cataract surgery and IOL power calculations POSTERIOR SEGMENT: Diabetic Retinopathy screening Age-Related Macular Degeneration Glaucoma screening and diagnosis Retinopathy of prematurity screening 9

10. #APPLICATIONS OF AI IN ANTERIOR SEGMENT DISEASES Wu X, Liu L, Zhao L, Guo C, Li R, Wang T, Yang X, et al. Application of artificial intelligence in anterior segment ophthalmic diseases: diversity and standardization. Ann Transl Med 2020;8:714 10

11. #AI IN KERATOCONUS SCREENING Preclinical detection of keratoconus- Bilateral data were effective in discriminating clinically normal fellow eyes of patients with keratoconus from normal eyes. Scheimpflug camera was used. The most sensitive parameters to date for early KC identification - combination of videokeratography , wavefront analysis, and optical coherence tomography. . 11

12. #Ai in detection of Amblyopia 12 1 st – universally visible 2 nd – visible with red green glass

13. #Ai in squint surgery 13

14. #AI IN REFRACTIVE SURGERY Refractive surgery has undergone rapid advancements in the last decades with good visual effects and long-term safety. Several refractive surgery types available both in the cornea and lens. The following surgeries with aid of laser are in vogue: 1)PRK, 2)LASIK 3)SMILE (Small incision lenticule extraction) more clinical data related to the surgery are being generated and more accuracy for the preoperative assessment and screening are required. Therefore, artificial intelligence assisting in diagnosis and surgery procedures may be needed. Kim TI, Alio Del Barrio JL, Wilkins M, Cochener B, Ang M. Refractive surgery. Lancet. 2019;393(10185):2085–98. https://doi.org/10.1016/ S0140-6736( 18)33209-4. 14

15. DEEP LEARNING IN PRE-OP ASSESSMENT IN REFRACTIVE SURGERY A. Grzybowski (ed.), Artificial Intelligence in Ophthalmology, © Springer Nature Switzerland AG 2021 https://doi.org/10.1007/978-3-030-78601-4_17 15

16. Yoo TK, Ryu IH, Lee G, et al . Adopting machine learning to automatically identify candidate patients for corneal refractive surgery. NPJ Digit Med 2019;2:59. 16

17. #AI based iol SAV-IOL (Swiss Advanced Vision) - Mimicking the human lens, the autofocus system sends signals that trigger micro-pumps to alter the curvature of the optical membrane through liquid displacement - in real time; in fact, the process occurs at 0.2 seconds, a speed equivalent to the accommodation of a healthy human eye. 17

18. Algorithm for iol power calculation Ladas Super Formula - Combinations of existing formulas (Hoffer Q. Holladay-1, Holladay-l with Koch adjust - ment , Haigis , and SRK/T formulas) and plotted into a 5-D surface Hill-Radial Basis Function (RBF) method - Based on Haag- Streit LENSTAR optical biometer Kane formula- Cloud-based formula Hill W. Hill-RBF Formula 3.0 [Internet]. Hill-RBF Calculator Version 3.0. https://rbfcalculator.com/ . Accessed 3 Feb 2021. Ladas JG, Siddiqui AA, Devgan U, Jun AS. A 3-D “Super Surface” combining modern intraocular lens formulas to generate a “Super Formula” and maximize accuracy. JAMA Ophthalmol. 2015;133(12):1431–6. Gatinel D, Debellemanière G, Saad A, Dubois M, Rampat R. Determining the theoretical effective lens position of thick intraocular lenses for machine learning based IOL power calculation and simulation. Transl Vis Sci Technol . 2021. 18

19. Break the limitations 19

20. Future of ai in cataract screening Hand-held retinal cameras,or slit lamp adapters attached to smartphones, can potentially provide better outreach for cataract screening,especially in rural or less-resourced areas. DR screening programs with automated cataract assessment based on retinal photos will reduce the additional cost. Recent work has demonstrated Al's capability in recognizing different phases of cataract surgery. This system may be particularly useful when training ophthalmology residents. 20

21. #Cataract Cataract is the leading cause of visual impairment worldwide, accounting for 62.5 million cases of visual impairment and blindness globally Screening: Cataracts are clinically diagnosed using slit-lamp biomicroscopy , and graded based on established clinical scales such as the Lens Opacities Classification System III Limitations: Manual" process requires clinical expertise Shortage of trained ophthalmologists Grading results are compromised by inter-examiner variability 21

22. Algorithm based on slit lamp photo Active Shape Model (ASM )- Identify the location of the crystalline lens and its nucleus on 5820 slit lamp photographs The ASM achieved 95% success rate in correctly identifying the location of lens. Cloud based AI- CC-Cruise- Diagnose and grade paediatric cataracts 22

23. 23

24. Algorithm based on slit lamp photo Residual neural network (Res Net) - 3-step sequential Al algorithm for the diagnosis and referral of cataracts Al system - first differentiate slit lamp photographs between mydriatic and nonmydriatic images, and between optical section and diffuse slit-lamp illumination. The images would be categorized as either normal ( ie , no cataract), cataractous, or postoperative IOL Type and severity of the cataract/posterior capsular o pacification -evaluated based on the Lens Opacities Classification System Il scale 24

25. Algorithim based on fundus photography ResNet-18 and ResNet-50 - visibility" of fundus images were used to denote four classes of cataract severity ( noncataract , mild, moderate, and severe cataract) 25

26. What to look for in selecting algoritm Number of Images in the training set How was the ground truth ascertained Whether the algorithm was trained on mydriatic or non mydriatic image Whether the algorithm was trained on single field or multiple field (field-investigations) 26 Selecting algorithm depends on circumstances of intended screening Diabetic Retinopathy is the main specific complication which affects 1/3 rd of the diabetic population and is a sight-threatening condition . * #AI IN DIABETIC RETINOPATHY

27. Final outcome in AI based system Referable DR, Non-referable DR, Ungradable image. Hospital referral (Referable DR + Ungradable image) Vision threatning DR Challenges Regional variations in outcomes Intergrader variations Agreement between human graders and the algorithm 27

28. Establishment of AI Algorithm 28 Abstraction of the proposed algorithmic pipeline Fundus heatmap overlaid on a fundus image

29. Commercial product FOR DR SCREENING 29

30. #AI IN DIABETIC RETINOPATHY In April 2018, the US Food and Drug Administration (FDA) approved an AI algorithm, developed by IDx , used with Topcon Fundus camera (Topcon Medical) for DR identification. a study was done on 900‑subjects in a primary‑care setting (10 primary care sites) with automated image analysis. Two 45‑degree digital images per eye (one centered on the macula , one centered on the optic nerve ) were obtained and analyzed. These images were compared with the stereo, widefield fundus imaging interpreted by the Wisconsin Fundus Photograph Reading Centre (FPRC) Saeedi P, Petersohn I, Salpea P, Malanda B, Karuranga S, Unwin N, Colagiuri S, Guariguata L, Motala AA, Ogurtsova K, Shaw JE, Bright D, Williams R; IDF Diabetes Atlas Committee. Global and regional diabetes prevalence estimates for 2019 and projections for 2030 and 2045: Results from the International Diabetes Federation Diabetes Atlas, 9 th  edition. Diabetes Res Clin Pract . 2019 Nov;157:107843. doi : 10.1016/j.diabres.2019.107843. Epub 2019 Sep 10. PMID: 31518657.* US Food and Drug Administration. FDA permits marketing of artificial intelligence‑based device to detect certain diabetes‑related eye problems. Available from: https://www.fda.gov/NewsEvents/ Newsroom/ PressAnnouncements /ucm604357.htm. Published April 11, 2018. [Last accessed on 2018 Aug 12]. 30

31. AI IN DIABETIC RETINOPATHY By autonomous comparison software provides one of the two results:* (1) If more than mild DR detected, refer to an eyecare professional (ECP); (2) If the results are negative for more than mild DR, rescreen in 12 months. Based on the analysis a new entity called more than minimal DR ( mtmDR ) was defined- the presence of ETDRS level 35 or higher (microaneurysms plus hard exudates, cotton wool spots, and/or mild retinal hemorrhages) and/or DME in at least one eye ** Sensitivity and specificity of the technology was 87.4% and 89.5% respectively for detecting more than mild DR Anti-VEGF outcome prediction and dose optimization in DME. Abramoff M. Artificial intelligence for automated detection of diabetic retinopathy in primary care. Paper presented at: Macular Society; February 22, 2018; Beverly Hills, CA. Available from: http:// webeye.ophth.uiowa.edu/ abramoff /MDA MacSocAbst 2018 02 22. Pdf [Internet]. [Last accessed on 2019 Mar 26].* Pros and Cons of Using an AI‑Based Diagnosis for Diabetic Retinopathy: Page 4 of 5 N.d. Optometry Times. Available from: http://www.optometrytimes.com/article/pros‑and‑cons-using‑ai‑b ased ‑diagnosis‑diabetic‑retinopathy. [Last accessed on 2018 Oct 29** 31

32. Idx-dr analysis Report 32

33. 33

34. EyeArt Retmarker SENSITIVITY 94.7% for any retinopathy, 93.8% for referable retinopathy (human graded as either ungradable, maculopathy, preproliferative , or proliferative), 99.6% for proliferative retinopathy 73.0% for any retinopathy, 85.0% for referable retinopathy, 97.9% for proliferative retinopathy. SPECIFICITY 20% for any DR 52.3% for any DR Tufail A, Kapetanakis VV, Salas-Vega S, Egan C, Rudisill C, Owen CG, et al. An observational study to assess if automated diabetic retinopathy image assessment software can replace one or more steps of manual imaging grading and to determine their cost-effectiveness. Health Technol Assess. 2016;20(92):1–72 34

35. MEDIOS-AI FOR DR ON SMART PHONE A study looking into 900 adult subjects with diabetes in India, where five retinal specialists graded images taken with the Remidio mobile camera for any DR or rDR . This was later compared to the Medios AI software running offline on an Iphone 6 Medios AI achieved good results with sensitivity and specificity pair for any DR of 83.3% and 95.5% and for rDR 93% and 92.5% Sosale B, Aravind SR, Murthy H, Narayana S, Sharma U, SGV G, et al. Simple, mobile-based artificial intelligence algorithm in the detection of diabetic retinopathy (SMART) study. BMJ Open Diabetes Res Amp Care. 2020;8(1):e000892 35

36. Pupil: Are they different in Diabetic Retinopathy Baseline pupil diameter (BPD) decreased with increasing severity of diabetic retinopathy. Mean velocity of pupillary constriction (VPC ) decreased progressively with increasing severity of retinopathy. Mean velocity of pupil re dilatation (VPD ) as compared to the control group was significantly reduced in the no DR (p - 0.03), mild NPDR (p - 0.038), moderate NPDR (p - 0.05), PDR group (p -0.02). Pupillary dynamics are abnormal in early stages of diabetic retinopathy and progress with increasing retinopathy severity. 36

37. Predicting Diabetic neuropathy from retinal imageS Retinal sensitivity was checked with microperimeter Alterations in Retinal Function seen in DN Increase in foveal thickness Reduced RNFL thickness Retinal images from people with diabetes can be used to identify individuals with DN. Simultaneous screening of DN and DR: new possibilities in preventing microvascular complications of Diabetes 37

38. Predicting response (good/poor) based on inflammatory markers Uveitis macular edema : Annotation of inflammatory markers Deep learning algorithm to detect these markers Use this algorithm to differentiate b/w DME with UME. Use the algorithm to see a difference in responder & no responder to the treatment 38

39. When not to use Ai 39

40. FLUID INTELLIGENCE APP 40

41. #Ocular ultrasound based AI 41 Heat maps highlighting regions of abnormalities detected using the DLA

42. #AI IN AGE-RELATED MACULAR DEGENERATION (ARMD) Age-related macular degeneration (AMD) accounts for approximately 9% of global blindness and is the leading cause of visual loss in developed countries. * The number of people with AMD worldwide is projected to be 196 million in 2020, rising substantially to 288 million in 2040. ** There are two forms of late AMD: 1)neovascular AMD 2) atrophic AMD, defined by geographic atrophy (GA) Congdon N, O’Colmain B, Klaver CCW, et al. Causes and prevalence of visual impairment among adults in the United States. Arch Ophthalmol Chic Ill 1960. 2004;122(4):477–85. https://doi.org/10.1001/ archopht.122.4.477. Wong WL, Su X, Li X, et al. Global prevalence of age-related macular degeneration and disease burden projection for 2020 and 2040: a systematic review and meta-analysis. Lancet Glob Health. 2014;2(2):e106– 16. https://doi.org/10.1016/S2214-109X ( 13)70145-1. 42

43. Establishment of Ai tool to screen ARMD ОСТ- Huge number of macular OCT's that are routinely done around the world Macular OCTs have dense three dimensional structural information that is usually consistently captured. OCTs provide structural detail that is not easily visible using conventional imaging techniques 43 Examples of identification of pathology by deep learning algorithm

44. Diagnosis of armd with ai 44 interrupted outer retina (pink) Interrupted RPE (lake blue) absence of outer retina (yellow ) absence of RPE (dark blue) hypertransmission «250um (red) hypertransmission z250um (green) Wei, W., Southern, J., Zhu, K. et al. Deep learning to detect macular atrophy in wet age-related macular degeneration using optical coherence tomography. Sci Rep 13, 8296 (2023). https:// doi.org /10.1038/s41598-023-35414-y

45. Deep learning network for armd 45

46. Commercially available DL system used in Armd AlexNet Google Net VGG Inception-V5 ResNet Inception-ResNet-V2 46

47. #Ai IN GLAUCOMA DIAGNOSIS Glaucoma is a leading cause of irreversible blindness, with a global prevalence of 3.5% and a global burden of 76 million affected people in 2020. Early detection and treatment can preserve vision in affected individuals. However, glaucoma is asymptomatic in early stages, as visual fi elds are not affected until 20 – 50% of corresponding retinal ganglion cells are lost. Tham YC, Li X, Wong TY, Quigley HA, Aung T, Cheng CY. Global prevalence of glaucoma and projections of glaucoma burden through 2040: a systematic review and meta-analysis. Ophthalmology. 2014;121:2081 – 90. Harwerth RS, Carter-Dawson L, Shen F, Smith EL 3rd, Crawford ML. Ganglion cell losses underlying visual fi eld defects from experimental glaucoma. Invest Ophthalmol Vis Sci. 1999; 40:2242 – 50. Harwerth RS, Carter-Dawson L, Smith EL 3rd, Barnes G, Holt WF, Crawford ML. Neural losses correlated with visual losses in clinical perimetry. Invest Ophthalmol Vis Sci. 2004;45:3152 – 60. 47

48. DEEP LEARNING AND DETECTION OF THE GLAUCOMATOUS DISC Assessment of optic nerve head (ONH) integrity is the foundation for detecting glaucomatous damage. PROCEDURE: Eileen L. Mayro et al; The impact of arti fi cial intelligence in the diagnosis and management of glaucoma; Eye (2020) 34:1 – 11 48

49. RNFL thickness maps extracted from all participant SS-OCT scans, Structural RNFL features were identified using principal component analysis (PCA) 49 En face and RNFL thickness map Bright area suggestive of RNFL defect An accuracy of 93% (based on expert opinion) was reported for a hybrid deep learning ANN analyzing single-scan SS-OCT data to classify eyes as normal or glaucomatous

50. Ai in glaucoma- Importance of Visual Field 50

51. AI tool predict treatment response 51 Improvement in visual field

52. Filtered forecasting method 52 Kalman filtering - Novel glaucoma forecasting tool that can generate a menu of personalized and dynamically updated target iOPs Fast progression Slow progression

53. TOWARDS “AUTOMATED GONIOSCOPY” An independent test set of 39 936 SS-OCT scans from 312 phakic subjects (128 SS-OCT meridional scans per eye) was analysed . Participants above 50 years with no previous history of intraocular surgery were consecutively recruited from glaucoma clinics. Indentation gonioscopy and dark room SS-OCT were performed. For each subject, all images were analysed by a DL-based network based on the VGG-16 architecture, for gonioscopic angle-closure detection. RESULTS: the AUC of the DLA was 0.85 (95% CI:0.80 to 0.90, with sensitivity of 83% and a specificity of 87%) to classify gonioscopic angle closure with the optimal cut-off value of >35% of circumferential angle closure. Porporato N, et al . Br J Ophthalmol 2022; 106 :1387–1392. doi:10.1136/bjophthalmol-2020-318275 53

54. COMBINING STRUCTURE AND FUNCTION IN GLAUCOMA DIAGNOSIS Global VF indices (mean defect, corrected loss variance, and short-term fluctuation) in combination with structural data (CDR, rim area, cup volume, and nerve fiber layer height) analyzed by an ANN was capable to correctly identify glaucomatous eyes with an accuracy of 88% in an early study. * This figure was higher than that of the same ANN trained with only structural or functional data. Brigatti L, Hoffman D, Caprioli J. Neural networks to identify glaucoma with structural and functional measurements. Am J Ophthalmol . 1996;121:511–21.* Bowd C, Hao J, Tavares IM, et al. Bayesian machine learning classifiers for combining structural and functional measurements to classify healthy and glaucomatous eyes . Invest Ophthalmol Vis Sci . 2008;49:945–53. 54

55. #AI IN ROP SCREENING IN NEWBORNS Retinopathy of Prematurity (ROP) accounts for 6–18% of childhood blindness, worldwide The decision to treat is primarily based on the presence of plus disease, defined as dilation and tortuosity of retinal vessels. However, clinical diagnosis of plus disease is highly subjective and variable. Chiang et al; International Classification of Retinopathy of Prematurity; Third Edition; Elsevier on behalf of AAO; 2021 JamesM . Brown et al; Automated Diagnosis of Plus Disease in Retinopathy of Prematurity Using Deep Convolutional Neural Networks; JAMA Ophthalmol . 2018;136(7):803-810. doi:10.1001/jamaophthalmol.2018.1934 55

56. Challenges to delivery of care: Clinical diagnosis is highly variable, and high interobserver inconsistency on plus disease diagnosis, even among ROP experts. The number of ophthalmologists and neonatologists willing and able to manage ROP is insufficient because of logistical difficulties, extensive training process, time-consuming examination, and significant malpractice liability. The incidence of ROP worldwide is rising because of advances in neonatology These challenges have stimulated research in developing quantitative and objective approaches to ROP diagnosis using computer-based image analysis (CBIA) 56

57. Commercially available DL system used in ROP • I-ROP DL algorithm Limitations Convolutional neural networks are only as robust as the data on which they are trained. System currently only classifies plus discase Ideally, a fully automated ROP screening platform could classify zone, stage, and overall disease category as well as predict need for treatment. JamesM . Brown et al; Automated Diagnosis of Plus Disease in Retinopathy of Prematurity Using Deep Convolutional Neural Networks; JAMA Ophthalmol . 2018;136(7):803-810. doi:10.1001/jamaophthalmol.2018.1934 57

58. ADVANTAGES/PROMISES OF AI IN OPHTHALMOLOGY outperform doctors, help to diagnose what is presently undiagnosable, help to treat what is presently untreatable, to recognize on images what is presently unrecognizable, predict the unpredictable, classify the unclassifiable, decrease the workflow inefficiencies, decrease hospital admissions and readmissions, increase medication adherence decrease patient harm decrease or eliminate misdiagnosis Provide personalised, high-precision medicine (“Pharmacogenomics”) Topol E. Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again. Basic Books, New York 2019 58

59. Challenges in the application of Ai Requires a large amount of data -Insufficient data may decrease the performance of DL models. For rare eye diseases - Low incidence makes it difficult for researchers to collect enough data for Al research. Requirement of prospective data -Some cohort studies have high requirements for patient data, and researchers need to collect data prospectively, which is very time- and energy-consuming 59

60. Security and Ethics Reporting guidelines- Al often overlooks clinically relevant details. For example, the criteria for participant recruitment, demographics, risk control. Sandardized reporting protocols are the solutions, like- CON-SORT-AI, STARD-AI, SPIRIT-Al and TRIPOD Security Medical -Al research should be performed on multicenter datasets. Data transfers between research collaborators, especially for international collaborations, are often limited because of patient privacy and data security. Ethics - As Al is gradually being integrated in clinical practice and medical professionals are getting used to it, there is also a growing trend for an ethical framework to guide the real application of DI systems. 60

61. WHAT DOES AI HERALD FOR THE FUTURE? R etinal microvasculature analyses potentially predict CVD risk factors (e.g. blood pressure, diabetes), direct CVD events (e.g. CVD mortality), retinal features (e.g. retinal vessel calibre) and CVD biomarkers (e.g. coronary artery calcium score) by “Google Deep Mind Health”* Automated grading of cataracts Managing pediatric conditions such as refractive errors, congenital cataracts, detect strabismus, predicting future high myopia, and diagnosing reading disability To automatically detect leukocoria in children from a recreational smartphone or digital camera photographs Measuring inner and outer retinal layer thicknesses to predict the risk for Alzheimer’s disease ** OCULOPLASTY: Measuring referable blepharoptosis and also in ocular oncology High fidelity mobile holograms and Extended reality(XR) for remote health care through upcoming 6G!! Wong DY, Lam MC, Ran A, Cheung CY. Artificial intelligence in retinal imaging for cardiovascular disease prediction: current trends and future directions. Current Opinion in Ophthalmology. 2022 Sep 1;33(5):440-6. Balyen L, Peto T. Promising artificial intelligence‑machine learning‑deep learning algorithms in ophthalmology. Asia Pac J Ophthalmol 2019;8:264‑72** 61

62. SUMMARY Artificial intelligence is a disruptive technology which has myriad applications at present and in the offing for the medical field. It has revolutionised Ophthalmology through aiding in screening and diagnosis of regularly encountered diseases like Keratoconus, cataract, DR, AMD, ROP, Glaucoma, etc. It has infused accuracy in pre- and post-op calculations in cataract and refractive surgeries. Prediction of systemic diseases through ocular measurements with AI is expected in the future. AI-is a double-edged sword- caution needs to be exercised by including the supervisory human touch. 62

63. 63