This content is only available for ASCRS Members
This content from the 2020 ASCRS Virtual Annual Meeting is only available to ASCRS members. To log in, click the teal "Login" button in the upper right-hand corner of this page.
Papers in this Session
Expand each tab below to view the paper abstract for each paper within this session.
Region of Interest Densitometry Analysis of DMEK Dehiscence on Anterior Segment Optical Coherence Tomography
Authors
Albert Y. Cheung, MD, ABO
Andrew G. Kalina
Alex B Im
Medi Eslani, MD
Elizabeth Yeu, MD
Purpose
To evaluate Anterior Segment Optical Coherence Tomography (AS-OCT, Heidelberg Spectralis) corneal densitometry (CD) in DMEK dehiscence
Methods
Retrospective chart review of eyes that underwent DMEK for Fuchs dystrophy between January 2018 to April 2019. Inclusion criteria were (1) documented areas of DMEK dehiscence on AS-OCT with (2) image quality index ≥25 and ≤34. Image analysis with ImageJ (NIH) compared total area (stroma above the dehiscence), mean stromal CD, and ratio of anterior-to-posterior (A:P) stromal CD for regions of DMEK dehiscence compared to the contralateral side with an attached DMEK graft. Control regions (with no dehiscence) and post-dehiscence resolution images were also analyzed.
Results
A total of 56 sectors of the 16 images from 7 eyes with DMEK dehiscence were included. Compared to the contralateral side, regions of DMEK dehiscence had larger total areas (<0.0001), lower mean stromal CD (p=0.0007), and higher A:P stromal CD (<0.0001). All control regions and post-dehiscence resolution images did not show any significant differences compared to the contralateral sides.
Conclusion
This novel technique to analyze AS-OCT can be useful to evaluate CD of specific regions of corneal pathology. Lower mean stromal CD and higher A:P stromal CD may specify corneal edema.
Albert Y. Cheung, MD, ABO
Andrew G. Kalina
Alex B Im
Medi Eslani, MD
Elizabeth Yeu, MD
Purpose
To evaluate Anterior Segment Optical Coherence Tomography (AS-OCT, Heidelberg Spectralis) corneal densitometry (CD) in DMEK dehiscence
Methods
Retrospective chart review of eyes that underwent DMEK for Fuchs dystrophy between January 2018 to April 2019. Inclusion criteria were (1) documented areas of DMEK dehiscence on AS-OCT with (2) image quality index ≥25 and ≤34. Image analysis with ImageJ (NIH) compared total area (stroma above the dehiscence), mean stromal CD, and ratio of anterior-to-posterior (A:P) stromal CD for regions of DMEK dehiscence compared to the contralateral side with an attached DMEK graft. Control regions (with no dehiscence) and post-dehiscence resolution images were also analyzed.
Results
A total of 56 sectors of the 16 images from 7 eyes with DMEK dehiscence were included. Compared to the contralateral side, regions of DMEK dehiscence had larger total areas (<0.0001), lower mean stromal CD (p=0.0007), and higher A:P stromal CD (<0.0001). All control regions and post-dehiscence resolution images did not show any significant differences compared to the contralateral sides.
Conclusion
This novel technique to analyze AS-OCT can be useful to evaluate CD of specific regions of corneal pathology. Lower mean stromal CD and higher A:P stromal CD may specify corneal edema.
Agreement in Central Corneal Thickness Measurement between Corvis ST and Ocular Response Analyzer
Authors
Reza Razeghinejad, MD
Ramin Salouti, MD
Nasrin Masihpour, MD
Maryam Ghoreyshi, MD
Hossein Nowroozzadeh, MD
Purpose
To evaluate the agreement and interchangeability of non-contact devices of Ocular Response Analyzer (ORA) and the Corvis ST in central corneal thickness measurement.
Methods
In this prospective comparative study, central corneal thickness (CCT) of both eyes of 284 healthy refractive surgery candidates without any prior ocular surgery were recruited. All patients had a complete ocular examination and corneal thickness measurement with both ORA and Corvis ST in the same session. Only the data from right eyes were analyzed. The mean values of CCT measurements by the ORA and Corvis ST were compared using paired t-test. Bland-Altman plots and 95% limits of agreement (95% LOA) were used to assess the agreement between the readings of the two devices.
Results
The mean (±SD) age of the subjects was 28.0 ± 4.9 years (range, 20 to 44), and 104 cases (36.6%) were male. The mean ORA CCT measurements was 555 ± 32 µm (range, 492 to 641) which was greater than that of the Corvis ST (536 ± 32 µm; range, 467 to 618 µm; mean difference, 19 ± 9 µm; P < 0.001). In only 7 eyes (2.4%), ORA readings were not greater than that of the Corvis ST. The two devices measurements showed strong correlation (Pearson correlation coefficient: 0.964; P < 0.001). The 95% LOA between the ORA and the Corvis ST was 2.2 to 35.9 µm. The 95% LOAs between the two devices were not significantly different for CCTs < 540 µm versus CCTs ≥ 540 µm.
Conclusion
The ORA CCT readings was up to 35.9 µm greater than Corvis ST in healthy subjects without prior intraocular surgeries. These two devices may not be used interchangeably for measuring the CCT.
Reza Razeghinejad, MD
Ramin Salouti, MD
Nasrin Masihpour, MD
Maryam Ghoreyshi, MD
Hossein Nowroozzadeh, MD
Purpose
To evaluate the agreement and interchangeability of non-contact devices of Ocular Response Analyzer (ORA) and the Corvis ST in central corneal thickness measurement.
Methods
In this prospective comparative study, central corneal thickness (CCT) of both eyes of 284 healthy refractive surgery candidates without any prior ocular surgery were recruited. All patients had a complete ocular examination and corneal thickness measurement with both ORA and Corvis ST in the same session. Only the data from right eyes were analyzed. The mean values of CCT measurements by the ORA and Corvis ST were compared using paired t-test. Bland-Altman plots and 95% limits of agreement (95% LOA) were used to assess the agreement between the readings of the two devices.
Results
The mean (±SD) age of the subjects was 28.0 ± 4.9 years (range, 20 to 44), and 104 cases (36.6%) were male. The mean ORA CCT measurements was 555 ± 32 µm (range, 492 to 641) which was greater than that of the Corvis ST (536 ± 32 µm; range, 467 to 618 µm; mean difference, 19 ± 9 µm; P < 0.001). In only 7 eyes (2.4%), ORA readings were not greater than that of the Corvis ST. The two devices measurements showed strong correlation (Pearson correlation coefficient: 0.964; P < 0.001). The 95% LOA between the ORA and the Corvis ST was 2.2 to 35.9 µm. The 95% LOAs between the two devices were not significantly different for CCTs < 540 µm versus CCTs ≥ 540 µm.
Conclusion
The ORA CCT readings was up to 35.9 µm greater than Corvis ST in healthy subjects without prior intraocular surgeries. These two devices may not be used interchangeably for measuring the CCT.
A Software Program to Standardize Diagnosis of Dry Eye
Authors
Ahmed R. Al-Ghoul, MD, ABO
Arman S. Grewal
Shakeel Qazi, BScOptom
Deepinder K. Dhaliwal, MD
Purpose
This purpose of the study is to evaluate a cloud-based dry eye software used to standardize diagnosis and treatment of dry eye and ensure consistency and continuity of care.
Methods
Six patients were examined by two doctors separately using the cloud-based software (CSI Dry Eye). The software requests doctors to input the following variables into the patient encounter (Medical History, Amount of “Itching”, Amount of Tearing, Symptoms scored using the Dry Eye Questionnaire, Fluorescein Staining, Osmolarity, Invasive Tear Breakup Time). Once the variables are entered, proprietary algorithms are used to classify severity of each variable. Doctors are then asked to verify which pattern they feel the patient falls under (from evaporative to aqueous, mild to severe) out of 9 general categories of dry eye disease.
Results
The results showed that out of six patients who were independently evaluated by two doctors, there was 83.3% agreement between the two doctors in categorizing the dry eye disease pattern using the dry eye software.
Conclusion
Preliminary data shows that the dry eye software is 83.3% effective in consistently categorizing dry eye disease in patients.
Ahmed R. Al-Ghoul, MD, ABO
Arman S. Grewal
Shakeel Qazi, BScOptom
Deepinder K. Dhaliwal, MD
Purpose
This purpose of the study is to evaluate a cloud-based dry eye software used to standardize diagnosis and treatment of dry eye and ensure consistency and continuity of care.
Methods
Six patients were examined by two doctors separately using the cloud-based software (CSI Dry Eye). The software requests doctors to input the following variables into the patient encounter (Medical History, Amount of “Itching”, Amount of Tearing, Symptoms scored using the Dry Eye Questionnaire, Fluorescein Staining, Osmolarity, Invasive Tear Breakup Time). Once the variables are entered, proprietary algorithms are used to classify severity of each variable. Doctors are then asked to verify which pattern they feel the patient falls under (from evaporative to aqueous, mild to severe) out of 9 general categories of dry eye disease.
Results
The results showed that out of six patients who were independently evaluated by two doctors, there was 83.3% agreement between the two doctors in categorizing the dry eye disease pattern using the dry eye software.
Conclusion
Preliminary data shows that the dry eye software is 83.3% effective in consistently categorizing dry eye disease in patients.
Artificial Intelligence - Anterior Segment Photos Computer Diagnositic
Author
Sunita R. Agarwal, MD
Purpose
To identify ocular and systemic diseases using camera on mobile application along with computational algorithms that can update themselves using artificial intelligence and the cloud
Methods
over 10000 patients data with anterior segment photographs and diagnosis is used to take measurements for physiological and pathological conditions. These are captured with computational algorithms where data is used as the raw data. Test subject can now be photographed along with the software application making it possible for the computer using artificial intelligence to give diagnosis ocular and systemic.
Results
The test image is compared with data image thus making disease identification possible along with computational software for a statistical study to confirm the diagnosis
Conclusion
Test subject diagnosis will be given by the software loaded on a mobile application this will be further enhanced by the individual doctors giving their diagnosis and all of this will be uploaded to the cloud. Increased use of the mobile camera diagnosis system will update itself using artificial intelligence.
Sunita R. Agarwal, MD
Purpose
To identify ocular and systemic diseases using camera on mobile application along with computational algorithms that can update themselves using artificial intelligence and the cloud
Methods
over 10000 patients data with anterior segment photographs and diagnosis is used to take measurements for physiological and pathological conditions. These are captured with computational algorithms where data is used as the raw data. Test subject can now be photographed along with the software application making it possible for the computer using artificial intelligence to give diagnosis ocular and systemic.
Results
The test image is compared with data image thus making disease identification possible along with computational software for a statistical study to confirm the diagnosis
Conclusion
Test subject diagnosis will be given by the software loaded on a mobile application this will be further enhanced by the individual doctors giving their diagnosis and all of this will be uploaded to the cloud. Increased use of the mobile camera diagnosis system will update itself using artificial intelligence.
A Prospective Study for Autonomous Diagnosis of Dry Eye Syndrome Using an Artificial Intelligence Algorithm
Authors
Collin B. Chase
Amr Elsawy, MSc
Taher K. Eleiwa, MD, MSc
Eyup Ozcan, MD
Mohamed Tolba, MD, MSc
Mohamed Abou Shousha, MD
Purpose
To validate the use of a multi-disease artificial intelligence deep learning algorithm in the diagnosis of dry eye syndrome (DES) using anterior segment optical coherence tomography (ASOCT).
Methods
In this prospective case study, 60 eyes of 30 patients (32 eyes of 16 patients with DES and 28 eyes of 14 healthy patients) were evaluated using a deep learning algorithm, Bascom Palmer AI (version 1.0, Miami, FL). A clinical diagnosis of DES or healthy was assigned to each patient based on dry eye testing performed by a researcher or ophthalmologist. The ASOCT scans were obtained using the Envisu R2210, Bioptigen (Buffalo Grove, IL, USA). ASOCT images were processed using the AI algorithm, capable of producing a diagnosis of healthy, dry eye, Fuchs’ dystrophy, and keratoconus. Prediction scores for diagnosis of dry eye syndrome and receiver operating curves (ROC) were generated.
Results
The deep learning algorithm was able to correctly differentiate all patients with dry eye from healthy subjects, thus achieving accuracy, sensitivity, and specificity of 100% at the patient level. The algorithm was able to correctly diagnose 29 out of 32 eyes with clinically diagnosed dry eye syndrome and 25 out of 28 healthy eyes. For the dry eye category, the area under the curve (AUC) was 0.993, with a sensitivity of 96.9% and a specificity of 95.8%. Within the dry eye category, the mean dry eye prediction score was 0.83±0.22, whereas within the healthy category, the mean dry eye prediction score was 0.11±0.17 (p-value<0.001).
Conclusion
The multi-disease predicting deep learning algorithm was effective at correctly diagnosing eyes with DES versus healthy eyes. Given the difficulty of diagnosing dry eye disease in a clinical setting without conducting multiple time-consuming dry eye tests, a deep learning algorithm with the ability to detect dry eye disease may be helpful.
Collin B. Chase
Amr Elsawy, MSc
Taher K. Eleiwa, MD, MSc
Eyup Ozcan, MD
Mohamed Tolba, MD, MSc
Mohamed Abou Shousha, MD
Purpose
To validate the use of a multi-disease artificial intelligence deep learning algorithm in the diagnosis of dry eye syndrome (DES) using anterior segment optical coherence tomography (ASOCT).
Methods
In this prospective case study, 60 eyes of 30 patients (32 eyes of 16 patients with DES and 28 eyes of 14 healthy patients) were evaluated using a deep learning algorithm, Bascom Palmer AI (version 1.0, Miami, FL). A clinical diagnosis of DES or healthy was assigned to each patient based on dry eye testing performed by a researcher or ophthalmologist. The ASOCT scans were obtained using the Envisu R2210, Bioptigen (Buffalo Grove, IL, USA). ASOCT images were processed using the AI algorithm, capable of producing a diagnosis of healthy, dry eye, Fuchs’ dystrophy, and keratoconus. Prediction scores for diagnosis of dry eye syndrome and receiver operating curves (ROC) were generated.
Results
The deep learning algorithm was able to correctly differentiate all patients with dry eye from healthy subjects, thus achieving accuracy, sensitivity, and specificity of 100% at the patient level. The algorithm was able to correctly diagnose 29 out of 32 eyes with clinically diagnosed dry eye syndrome and 25 out of 28 healthy eyes. For the dry eye category, the area under the curve (AUC) was 0.993, with a sensitivity of 96.9% and a specificity of 95.8%. Within the dry eye category, the mean dry eye prediction score was 0.83±0.22, whereas within the healthy category, the mean dry eye prediction score was 0.11±0.17 (p-value<0.001).
Conclusion
The multi-disease predicting deep learning algorithm was effective at correctly diagnosing eyes with DES versus healthy eyes. Given the difficulty of diagnosing dry eye disease in a clinical setting without conducting multiple time-consuming dry eye tests, a deep learning algorithm with the ability to detect dry eye disease may be helpful.
Diagnostic Performance of an Artificial Intelligence Algorithm in Fuchs Endothelial Cell Dystrophy
Authors
Taher K. Eleiwa, MD, MSc
Amr Elsawy, MSc
Collin B. Chase
Eyup Ozcan, MD
Mohamed Tolba, MD, MSc
Mohamed Abou Shousha, MD
Purpose
To prospectively evaluate the diagnostic performance of a multi-disease deep learning-based artificial intelligence (AI) algorithm to autonomously diagnose Fuchs endothelial cell dystrophy (FECD) using anterior segment optical coherence tomography (AS-OCT).
Methods
In this prospective case-control study, 41 eyes of 24 subjects (15 eyes of 10 FECD, and 28 eyes of 14 controls) were imaged using AS-OCT (Envisu R2210, Bioptigen, Buffalo Grove, IL, USA). Then, an AI algorithm (Bascom Palmer AI, version 1.0, Miami, FL) was used to discriminate FECD eyes using the captured AS-OCT scans. Hereby, a prediction score for FECD was determined and receiver operating characteristic (ROC) curve was generated.
Results
The AI algorithm was able to autonomously diagnose FECD with an area under the curve (AUC) of 0.999, a 100% sensitivity and 97.8% specificity. The mean FECD prediction score was 0.92 ± 0.26 for FECD, compared to 0.07 ± 0.18 in healthy controls (P <0.001).
Conclusion
With the presented AI deep learning-based algorithm, autonomous detection of FECD is possible with excellent accuracy, sensitivity and specificity.
Taher K. Eleiwa, MD, MSc
Amr Elsawy, MSc
Collin B. Chase
Eyup Ozcan, MD
Mohamed Tolba, MD, MSc
Mohamed Abou Shousha, MD
Purpose
To prospectively evaluate the diagnostic performance of a multi-disease deep learning-based artificial intelligence (AI) algorithm to autonomously diagnose Fuchs endothelial cell dystrophy (FECD) using anterior segment optical coherence tomography (AS-OCT).
Methods
In this prospective case-control study, 41 eyes of 24 subjects (15 eyes of 10 FECD, and 28 eyes of 14 controls) were imaged using AS-OCT (Envisu R2210, Bioptigen, Buffalo Grove, IL, USA). Then, an AI algorithm (Bascom Palmer AI, version 1.0, Miami, FL) was used to discriminate FECD eyes using the captured AS-OCT scans. Hereby, a prediction score for FECD was determined and receiver operating characteristic (ROC) curve was generated.
Results
The AI algorithm was able to autonomously diagnose FECD with an area under the curve (AUC) of 0.999, a 100% sensitivity and 97.8% specificity. The mean FECD prediction score was 0.92 ± 0.26 for FECD, compared to 0.07 ± 0.18 in healthy controls (P <0.001).
Conclusion
With the presented AI deep learning-based algorithm, autonomous detection of FECD is possible with excellent accuracy, sensitivity and specificity.
Objective Grading of Fuchs’ Endothelial Corneal Dystrophy Using Three-Dimensional Endothelium/Descemet’s Membrane Complex Thickness Map
Authors
Taher K. Eleiwa, MD, MSc
Amr Elsawy, MSc
Sonia H. Yoo, MD, ABO
Mohamed Abou Shousha, MD
Methods
In this cross-sectional study, 79 eyes of 66 subjects (31 FECD patients, and 35 age- and gender-matched controls) were imaged using high definition optical coherence tomography (HD-OCT) device (Envisu R2210, Bioptigen, Buffalo Grove, IL, USA). Automatic segmental tomography was performed to generate 3D maps of total corneal thickness (TCT) and En/DMT. FECD was clinically classified into early-stage (without corneal edema) and late-stage (with corneal edema). En/DMT, TCT, and central to peripheral total corneal thickness ratio (CPTR) were evaluated and correlated to the clinical severity.
Results
En/DMT, CPTR, and regional TCT were significantly higher in FECD group compared to controls (P < 0.001). The peripheral En/DMT, paracentral En/DMT, central En/DMT, and CPTR provided excellent discrimination between FECD and controls (area under the curve, AUC, 0.99, 0.97, 0.96, and 0.931, respectively), whereas the central TCT and paracentral TCT achieved an AUC of 0.83 and 0.84, respectively. The central En/DMT, paracentral En/DMT, peripheral En/DMT, and CPTR were strongly correlated with the clinical stage of FECD (R= 0.879, P < 0.001; R= 0.902, P = 0.01; R= 0.898, P < 0.001; and R=0.721, P<0.001, respectively), compared to (R = 0.585, P < 0.001) for central TCT.
Conclusion
3D-En/DMT Map is a novel tool for the diagnosis of FECD that can be used to objectively quantify the disease severity.
Taher K. Eleiwa, MD, MSc
Amr Elsawy, MSc
Sonia H. Yoo, MD, ABO
Mohamed Abou Shousha, MD
Methods
In this cross-sectional study, 79 eyes of 66 subjects (31 FECD patients, and 35 age- and gender-matched controls) were imaged using high definition optical coherence tomography (HD-OCT) device (Envisu R2210, Bioptigen, Buffalo Grove, IL, USA). Automatic segmental tomography was performed to generate 3D maps of total corneal thickness (TCT) and En/DMT. FECD was clinically classified into early-stage (without corneal edema) and late-stage (with corneal edema). En/DMT, TCT, and central to peripheral total corneal thickness ratio (CPTR) were evaluated and correlated to the clinical severity.
Results
En/DMT, CPTR, and regional TCT were significantly higher in FECD group compared to controls (P < 0.001). The peripheral En/DMT, paracentral En/DMT, central En/DMT, and CPTR provided excellent discrimination between FECD and controls (area under the curve, AUC, 0.99, 0.97, 0.96, and 0.931, respectively), whereas the central TCT and paracentral TCT achieved an AUC of 0.83 and 0.84, respectively. The central En/DMT, paracentral En/DMT, peripheral En/DMT, and CPTR were strongly correlated with the clinical stage of FECD (R= 0.879, P < 0.001; R= 0.902, P = 0.01; R= 0.898, P < 0.001; and R=0.721, P<0.001, respectively), compared to (R = 0.585, P < 0.001) for central TCT.
Conclusion
3D-En/DMT Map is a novel tool for the diagnosis of FECD that can be used to objectively quantify the disease severity.