Yuri Fujino
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Research Interests
Yuri Fujino has a total of 59 co-authors. Here are the top co-authors by number of shared publications:
Hiroshi Murata
Ophthalmology, National Center for Global Health and Medicine, Shinjuku-ku, Japan
22 shared publications- Ryo Asaoka
Affiliations Department of Ophthalmology, Seirei Hamamatsu General Hospital, Hamamatsu, Shizuoka, Japan, Seirei Christopher University, Hamamatsu, Shizuoka, Japan, The Graduate School for the Creation of New Photonics Industries, Shizuoka, Japan, Organization for Innovation and Social Collaboration, National University Corporation Shizuoka University, Hamamatsu, Shizuoka, Japan
21 shared publications - Masato Matsuura
Department of Ophthalmology, University of Tokyo, Tokyo, Japan
12 shared publications - Kazunori Hirasawa
Department of Ophthalmology, Kitasato University School of Medicine, Sagamihara, Japan
8 shared publications Kazuhiko Mori
Department of Ophthalmology, Kyoto Prefectural University of Medicine, Kyoto, Kyoto, Japan
7 shared publicationsMasato Matsuura
Department of Ophthalmology, Graduate School of Medicine and Faculty of Medicine, The University of Tokyo, Tokyo, 113-8655, Japan
6 shared publications
Latest Publications
- 19 October 2022
- Published by BMJ in British Journal of Ophthalmology
- Vol. 108  (1) , 78-83
Accuracy of pattern deviation in estimating the glaucomatous damage in the central 10° visual field in eyes with glaucoma and cataract
Background/aims: The accuracy of pattern deviation (PD) in estimating the damage to the glaucomatous visual field (VF) in the central 10° in eyes with glaucoma and cataract is unclear. Methods: This retrospective study includes 63 eyes of 52 glaucoma patients who successfully underwent cataract surgery or cataract surgery plus iStent implantation. Using the Humphrey Field Analyser 10–2 test, VF was measured within 6 months preoperatively and postoperatively (VF pre and VF post , respectively). The mean total deviation values in VF post (mTD post ) indicates glaucomatous damage without cataract and the difference between this value and mean PD values in VF pre (mPD pre ) was evaluated (εmPD). The effect of cataract was then evaluated as the difference between mTD post and mTD pre (ΔmTD), while the effects of mTD post and ΔmTD on εmPD were also assessed. In addition, based on preoperative visual acuity (VA pre ) and VF pre , the optimal model for predicting mTD post was identified. The error of this method (εOptimalModel) was estimated as the difference against mTD post , which was compared with εmPD. Results: Compared with mTD pre , there was a significant improvement in mTD post (p=0.028). A significant difference was observed between mPD pre and mTD post (ppost or ΔmTD (p<0.001 and p=0.0444, respectively). The absolute εOptimalModel was significantly smaller than the absolute εmPD (p<0.001). Conclusions: This study warns clinicians that PD of the central 10° VF might underestimate the glaucomatous VF damage with the progression of glaucoma and overestimate it as a cataract progresses.
- 19 October 2022
- Published by BMJ in British Journal of Ophthalmology
- Vol. 108  (1) , 78-83
Accuracy of pattern deviation in estimating the glaucomatous damage in the central 10° visual field in eyes with glaucoma and cataract
Background/aims: The accuracy of pattern deviation (PD) in estimating the damage to the glaucomatous visual field (VF) in the central 10° in eyes with glaucoma and cataract is unclear. Methods: This retrospective study includes 63 eyes of 52 glaucoma patients who successfully underwent cataract surgery or cataract surgery plus iStent implantation. Using the Humphrey Field Analyser 10–2 test, VF was measured within 6 months preoperatively and postoperatively (VF pre and VF post , respectively). The mean total deviation values in VF post (mTD post ) indicates glaucomatous damage without cataract and the difference between this value and mean PD values in VF pre (mPD pre ) was evaluated (εmPD). The effect of cataract was then evaluated as the difference between mTD post and mTD pre (ΔmTD), while the effects of mTD post and ΔmTD on εmPD were also assessed. In addition, based on preoperative visual acuity (VA pre ) and VF pre , the optimal model for predicting mTD post was identified. The error of this method (εOptimalModel) was estimated as the difference against mTD post , which was compared with εmPD. Results: Compared with mTD pre , there was a significant improvement in mTD post (p=0.028). A significant difference was observed between mPD pre and mTD post (ppost or ΔmTD (p<0.001 and p=0.0444, respectively). The absolute εOptimalModel was significantly smaller than the absolute εmPD (p<0.001). Conclusions: This study warns clinicians that PD of the central 10° VF might underestimate the glaucomatous VF damage with the progression of glaucoma and overestimate it as a cataract progresses.
- Open Access
- 18 November 2021
- Published by BMJ in BMJ Open Ophthalmology
- Vol. 6  (1) , e000900
Association between the number of visual fields and the accuracy of future prediction in eyes with retinitis pigmentosa
Purpose To evaluate the minimum number of visual fields (VFs) required to precisely predict future VFs in eyes with retinitis pigmentosa (RP). Methods A series of 12 VFs (Humphrey Field Analyzer 10–2 test (8.9 years in average) were analysed from 102 eyes of 52 patients with RP. The absolute error to predict the 12th VF using the prior 11 VFs was calculated in a pointwise manner, using the linear regression, and the 95% CI range was determined. Then, using 3–10 initial VFs, next VFs (4th to 11th VFs, respectively) were also predicted. The minimum number of VFs required for the mean absolute prediction error to reach the 95% CI was identified. Similar analyses were iterated for the second and third next VF predictions. Similar analyses were conducted using mean deviation (MD). Results In the pointwise analysis, the minimum number of VFs required to reach the 95% CI for the 12th VF was five (first and second next VF predictions) and six (third next VF prediction). For the MD analysis, three (first and second next VF predictions) and four (third next VF prediction) VFs were required to reach 95% CI for the 12th VF. Conclusions The minimum number of VFs required to obtain accurate predictions of the future VF was five or six in the pointwise analysis and three or four in the analysis with MD.
- 18 November 2021
- Published by BMJ in BMJ Open Ophthalmology
- Vol. 6  (1) , e000900
Open AccessAssociation between the number of visual fields and the accuracy of future prediction in eyes with retinitis pigmentosa
Purpose To evaluate the minimum number of visual fields (VFs) required to precisely predict future VFs in eyes with retinitis pigmentosa (RP). Methods A series of 12 VFs (Humphrey Field Analyzer 10–2 test (8.9 years in average) were analysed from 102 eyes of 52 patients with RP. The absolute error to predict the 12th VF using the prior 11 VFs was calculated in a pointwise manner, using the linear regression, and the 95% CI range was determined. Then, using 3–10 initial VFs, next VFs (4th to 11th VFs, respectively) were also predicted. The minimum number of VFs required for the mean absolute prediction error to reach the 95% CI was identified. Similar analyses were iterated for the second and third next VF predictions. Similar analyses were conducted using mean deviation (MD). Results In the pointwise analysis, the minimum number of VFs required to reach the 95% CI for the 12th VF was five (first and second next VF predictions) and six (third next VF prediction). For the MD analysis, three (first and second next VF predictions) and four (third next VF prediction) VFs were required to reach 95% CI for the 12th VF. Conclusions The minimum number of VFs required to obtain accurate predictions of the future VF was five or six in the pointwise analysis and three or four in the analysis with MD.
- 14 July 2021
- Published by BMJ in British Journal of Ophthalmology
- Vol. 106  (4) , 587-592
Deep learning-assisted (automatic) diagnosis of glaucoma using a smartphone
Background/aims To validate a deep learning algorithm to diagnose glaucoma from fundus photography obtained with a smartphone. Methods A training dataset consisting of 1364 colour fundus photographs with glaucomatous indications and 1768 colour fundus photographs without glaucomatous features was obtained using an ordinary fundus camera. The testing dataset consisted of 73 eyes of 73 patients with glaucoma and 89 eyes of 89 normative subjects. In the testing dataset, fundus photographs were acquired using an ordinary fundus camera and a smartphone. A deep learning algorithm was developed to diagnose glaucoma using a training dataset. The trained neural network was evaluated by prediction result of the diagnostic of glaucoma or normal over the test datasets, using images from both an ordinary fundus camera and a smartphone. Diagnostic accuracy was assessed using the area under the receiver operating characteristic curve (AROC). Results The AROC with a fundus camera was 98.9% and 84.2% with a smartphone. When validated only in eyes with advanced glaucoma (mean deviation value < −12 dB, N=26), the AROC with a fundus camera was 99.3% and 90.0% with a smartphone. There were significant differences between these AROC values using different cameras. Conclusion The usefulness of a deep learning algorithm to automatically screen for glaucoma from smartphone-based fundus photographs was validated. The algorithm had a considerable high diagnostic ability, particularly in eyes with advanced glaucoma.
- 14 July 2021
- Published by BMJ in British Journal of Ophthalmology
- Vol. 106  (4) , 587-592
Deep learning-assisted (automatic) diagnosis of glaucoma using a smartphone
Background/aims To validate a deep learning algorithm to diagnose glaucoma from fundus photography obtained with a smartphone. Methods A training dataset consisting of 1364 colour fundus photographs with glaucomatous indications and 1768 colour fundus photographs without glaucomatous features was obtained using an ordinary fundus camera. The testing dataset consisted of 73 eyes of 73 patients with glaucoma and 89 eyes of 89 normative subjects. In the testing dataset, fundus photographs were acquired using an ordinary fundus camera and a smartphone. A deep learning algorithm was developed to diagnose glaucoma using a training dataset. The trained neural network was evaluated by prediction result of the diagnostic of glaucoma or normal over the test datasets, using images from both an ordinary fundus camera and a smartphone. Diagnostic accuracy was assessed using the area under the receiver operating characteristic curve (AROC). Results The AROC with a fundus camera was 98.9% and 84.2% with a smartphone. When validated only in eyes with advanced glaucoma (mean deviation value < −12 dB, N=26), the AROC with a fundus camera was 99.3% and 90.0% with a smartphone. There were significant differences between these AROC values using different cameras. Conclusion The usefulness of a deep learning algorithm to automatically screen for glaucoma from smartphone-based fundus photographs was validated. The algorithm had a considerable high diagnostic ability, particularly in eyes with advanced glaucoma.
- 5 March 2021
- Published by BMJ in British Journal of Ophthalmology
- Vol. 106  (8) , 1098-1103
Investigating the clinical usefulness of definitions of progression with 10-2 visual field
Background/Aim To investigate the clinical validity of the Guided Progression Analysis definition (GPAD) and cluster-based definition (CBD) with the Humphrey Field Analyzer 10-2 test in diagnosing glaucomatous visual field (VF) progression, and to introduce a novel definition with optimised specificity by combining the ‘any-location’ and ‘cluster-based’ approaches (hybrid definition). Methods 64 400 stable glaucomatous VFs were simulated from 664 pairs of 10-2 tests (10 sets × 10 VF series × 664 eyes; data set 1). Using these simulated VFs, the specificity to detect progression and the effects of changing the parameters (number of test locations or consecutive VF tests, and percentile cut-off values) were investigated. The hybrid definition was designed as the combination where the specificity was closest to 95.0%. Subsequently, another 5000 actual glaucomatous 10-2 tests from 500 eyes (10 VFs each) were collected (data set 2), and their accuracy (sensitivity, specificity and false positive rate) and the time needed to detect VF progression were evaluated. Results The specificity values calculated using data set 1 with GPAD and CBD were 99.6% and 99.8%. Using data set 2, the hybrid definition had a higher sensitivity than GPAD and CBD, without detriment to the specificity or false positive rate. The hybrid definition also detected progression significantly earlier than GPAD and CBD (at 3.1 years vs 4.2 years and 4.1 years, respectively). Conclusions GPAD and CBD had specificities of 99.6% and 99.8%, respectively. A novel hybrid definition (with a specificity of 95.5%) had higher sensitivity and enabled earlier detection of progression.
- 5 March 2021
- Published by BMJ in British Journal of Ophthalmology
- Vol. 106  (8) , 1098-1103
Investigating the clinical usefulness of definitions of progression with 10-2 visual field
Background/Aim To investigate the clinical validity of the Guided Progression Analysis definition (GPAD) and cluster-based definition (CBD) with the Humphrey Field Analyzer 10-2 test in diagnosing glaucomatous visual field (VF) progression, and to introduce a novel definition with optimised specificity by combining the ‘any-location’ and ‘cluster-based’ approaches (hybrid definition). Methods 64 400 stable glaucomatous VFs were simulated from 664 pairs of 10-2 tests (10 sets × 10 VF series × 664 eyes; data set 1). Using these simulated VFs, the specificity to detect progression and the effects of changing the parameters (number of test locations or consecutive VF tests, and percentile cut-off values) were investigated. The hybrid definition was designed as the combination where the specificity was closest to 95.0%. Subsequently, another 5000 actual glaucomatous 10-2 tests from 500 eyes (10 VFs each) were collected (data set 2), and their accuracy (sensitivity, specificity and false positive rate) and the time needed to detect VF progression were evaluated. Results The specificity values calculated using data set 1 with GPAD and CBD were 99.6% and 99.8%. Using data set 2, the hybrid definition had a higher sensitivity than GPAD and CBD, without detriment to the specificity or false positive rate. The hybrid definition also detected progression significantly earlier than GPAD and CBD (at 3.1 years vs 4.2 years and 4.1 years, respectively). Conclusions GPAD and CBD had specificities of 99.6% and 99.8%, respectively. A novel hybrid definition (with a specificity of 95.5%) had higher sensitivity and enabled earlier detection of progression.
- Open Access
- 13 January 2021
- Published by BMJ in British Journal of Ophthalmology
- Vol. 106  (5) , 660-666
Comparing the usefulness of a new algorithm to measure visual field using the variational Bayes linear regression in glaucoma patients, in comparison to the Swedish interactive thresholding algorithm
Background/aims We previously reported that the visual field (VF) prediction model using the variational Bayes linear regression (VBLR) is useful for accurately predicting VF progression in glaucoma (Invest Ophthalmol Vis Sci. 2014, 2018). We constructed a VF measurement algorithm using VBLR, and the purpose of this study was to investigate its usefulness. Method 122 eyes of 73 patients with open-angle glaucoma were included in the current study. VF measurement was performed using the currently proposed VBLR programme with AP-7700 perimetry (KOWA). VF measurements were also conducted using the Swedish interactive thresholding algorithm (SITA) standard programme with Humphrey field analyser. VF measurements were performed using the 24–2 test grid. Visual sensitivities, test–retest reproducibility and measurement duration were compared between the two algorithms. Result Mean mean deviation (MD) values with SITA standard were −7.9 and −8.7 dB (first and second measurements), whereas those with VBLR-VF were −8.2 and −8.0 dB, respectively. There were no significant differences across these values. The correlation coefficient of MD values between the 2 algorithms was 0.97 or 0.98. Test–retest reproducibility did not differ between the two algorithms. Mean measurement duration with SITA standard was 6 min and 02 s or 6 min and 00 s (first or second measurement), whereas a significantly shorter duration was associated with VBLR-VF (5 min and 23 s or 5 min and 30 s). Conclusion VBLR-VF reduced test duration while maintaining the same accuracy as the SITA-standard.
- 13 January 2021
- Published by BMJ in British Journal of Ophthalmology
- Vol. 106  (5) , 660-666
Open AccessComparing the usefulness of a new algorithm to measure visual field using the variational Bayes linear regression in glaucoma patients, in comparison to the Swedish interactive thresholding algorithm
Background/aims We previously reported that the visual field (VF) prediction model using the variational Bayes linear regression (VBLR) is useful for accurately predicting VF progression in glaucoma (Invest Ophthalmol Vis Sci. 2014, 2018). We constructed a VF measurement algorithm using VBLR, and the purpose of this study was to investigate its usefulness. Method 122 eyes of 73 patients with open-angle glaucoma were included in the current study. VF measurement was performed using the currently proposed VBLR programme with AP-7700 perimetry (KOWA). VF measurements were also conducted using the Swedish interactive thresholding algorithm (SITA) standard programme with Humphrey field analyser. VF measurements were performed using the 24–2 test grid. Visual sensitivities, test–retest reproducibility and measurement duration were compared between the two algorithms. Result Mean mean deviation (MD) values with SITA standard were −7.9 and −8.7 dB (first and second measurements), whereas those with VBLR-VF were −8.2 and −8.0 dB, respectively. There were no significant differences across these values. The correlation coefficient of MD values between the 2 algorithms was 0.97 or 0.98. Test–retest reproducibility did not differ between the two algorithms. Mean measurement duration with SITA standard was 6 min and 02 s or 6 min and 00 s (first or second measurement), whereas a significantly shorter duration was associated with VBLR-VF (5 min and 23 s or 5 min and 30 s). Conclusion VBLR-VF reduced test duration while maintaining the same accuracy as the SITA-standard.
- 13 January 2021
- Published by BMJ in British Journal of Ophthalmology
- Vol. 106  (4) , 497-501
Validating the usefulness of sectorwise regression of visual field in the central 10°
Aim To evaluate the usefulness of the application of the clustering method to the trend analysis (sectorwise regression) in comparison with the pointwise linear regression (PLR). Methods This study included 153 eyes of 101 patients with open-angle glaucoma. With PLR, the total deviation (TD) values of the 10th visual field (VF) were predicted using the shorter VF sequences (from first 3 to 9) by extrapolating TD values against time in a pointwise manner. Then, 68 test points were stratified into 29 sectors. In each sector, the mean of TD values was calculated and allocated to all test points belonging to the sector. Subsequently, the TD values of the 10th VF were predicted by extrapolating the allocated TD value against time in a pointwise manner. Similar analyses were conducted to predict the 11th–16th VFs using the first 10 VFs. Results When predicting the 10th VF using the shorter sequences, the mean absolute error (MAE) values were significantly smaller in the sectorwise regression than in PLR. When predicting from the 11th and 16th VFs using the first 10 VFs, the MAE values were significantly larger in the sectorwise regression than in PLR when predicting the 11th VF; however, no significant difference was observed with other VF predictions. Conclusion Accurate prediction was achieved using the sectorwise regression, in particular when a small number of VFs were used in the prediction. The accuracy of the sectorwise regression was not hampered in longer follow-up compared with PLR.
- 13 January 2021
- Published by BMJ in British Journal of Ophthalmology
- Vol. 106  (4) , 497-501
Validating the usefulness of sectorwise regression of visual field in the central 10°
Aim To evaluate the usefulness of the application of the clustering method to the trend analysis (sectorwise regression) in comparison with the pointwise linear regression (PLR). Methods This study included 153 eyes of 101 patients with open-angle glaucoma. With PLR, the total deviation (TD) values of the 10th visual field (VF) were predicted using the shorter VF sequences (from first 3 to 9) by extrapolating TD values against time in a pointwise manner. Then, 68 test points were stratified into 29 sectors. In each sector, the mean of TD values was calculated and allocated to all test points belonging to the sector. Subsequently, the TD values of the 10th VF were predicted by extrapolating the allocated TD value against time in a pointwise manner. Similar analyses were conducted to predict the 11th–16th VFs using the first 10 VFs. Results When predicting the 10th VF using the shorter sequences, the mean absolute error (MAE) values were significantly smaller in the sectorwise regression than in PLR. When predicting from the 11th and 16th VFs using the first 10 VFs, the MAE values were significantly larger in the sectorwise regression than in PLR when predicting the 11th VF; however, no significant difference was observed with other VF predictions. Conclusion Accurate prediction was achieved using the sectorwise regression, in particular when a small number of VFs were used in the prediction. The accuracy of the sectorwise regression was not hampered in longer follow-up compared with PLR.
- 1 October 2020
- Published by Elsevier in American Journal of Ophthalmology
- Vol. 218, 304-313
Predicting the Glaucomatous Central 10-Degree Visual Field From Optical Coherence Tomography Using Deep Learning and Tensor Regression
- 1 October 2020
- Published by Elsevier in American Journal of Ophthalmology
- Vol. 218, 304-313
Predicting the Glaucomatous Central 10-Degree Visual Field From Optical Coherence Tomography Using Deep Learning and Tensor Regression
- 27 June 2020
- Published by BMJ in British Journal of Ophthalmology
- Vol. 105  (4) , 507-513
Deep learning model to predict visual field in central 10° from optical coherence tomography measurement in glaucoma
Background/Aim: To train and validate the prediction performance of the deep learning (DL) model to predict visual field (VF) in central 10° from spectral domain optical coherence tomography (SD-OCT). Methods: This multicentre, cross-sectional study included paired Humphrey field analyser (HFA) 10-2 VF and SD-OCT measurements from 591 eyes of 347 patients with open-angle glaucoma (OAG) or normal subjects for the training data set. We trained a convolutional neural network (CNN) for predicting VF threshold (TH) sensitivity values from the thickness of the three macular layers: retinal nerve fibre layer, ganglion cell layer+inner plexiform layer and outer segment+retinal pigment epithelium. We implemented pattern-based regularisation on top of CNN to avoid overfitting. Using an external testing data set of 160 eyes of 131 patients with OAG, the prediction performance (absolute error (AE) and R2 between predicted and actual TH values) was calculated for (1) mean TH in whole VF and (2) each TH of 68 points. For comparison, we trained support vector machine (SVM) and multiple linear regression (MLR). Results: AE of whole VF with CNN was 2.84±2.98 (mean±SD) dB, significantly smaller than those with SVM (5.65±5.12 dB) and MLR (6.96±5.38 dB) (all, p2 with CNN was 0.74 for the mean TH of whole VF, and 0.44±0.24 for the overall 68 points. Conclusion: DL model showed considerably accurate prediction of HFA 10-2 VF from SD-OCT.
- 27 June 2020
- Published by BMJ in British Journal of Ophthalmology
- Vol. 105  (4) , 507-513
Deep learning model to predict visual field in central 10° from optical coherence tomography measurement in glaucoma
Background/Aim: To train and validate the prediction performance of the deep learning (DL) model to predict visual field (VF) in central 10° from spectral domain optical coherence tomography (SD-OCT). Methods: This multicentre, cross-sectional study included paired Humphrey field analyser (HFA) 10-2 VF and SD-OCT measurements from 591 eyes of 347 patients with open-angle glaucoma (OAG) or normal subjects for the training data set. We trained a convolutional neural network (CNN) for predicting VF threshold (TH) sensitivity values from the thickness of the three macular layers: retinal nerve fibre layer, ganglion cell layer+inner plexiform layer and outer segment+retinal pigment epithelium. We implemented pattern-based regularisation on top of CNN to avoid overfitting. Using an external testing data set of 160 eyes of 131 patients with OAG, the prediction performance (absolute error (AE) and R2 between predicted and actual TH values) was calculated for (1) mean TH in whole VF and (2) each TH of 68 points. For comparison, we trained support vector machine (SVM) and multiple linear regression (MLR). Results: AE of whole VF with CNN was 2.84±2.98 (mean±SD) dB, significantly smaller than those with SVM (5.65±5.12 dB) and MLR (6.96±5.38 dB) (all, p2 with CNN was 0.74 for the mean TH of whole VF, and 0.44±0.24 for the overall 68 points. Conclusion: DL model showed considerably accurate prediction of HFA 10-2 VF from SD-OCT.
- Open Access
- 20 April 2020
- Published by Springer Nature in Scientific Reports
- Vol. 10  (1) , 1-8
Visualizing the dynamic change of Ocular Response Analyzer waveform using Variational Autoencoder in association with the peripapillary retinal arteries angle
The aim of the current study is to identify possible new Ocular Response Analyzer (ORA) waveform parameters related to changes of retinal structure/deformation, as measured by the peripapillary retinal arteries angle (PRAA), using a generative deep learning method of variational autoencoder (VAE). Fifty-four eyes of 52 subjects were enrolled. The PRAA was calculated from fundus photographs and was used to train a VAE model. By analyzing the ORA waveform reconstructed (noise filtered) using VAE, a novel ORA waveform parameter (Monot1-2), was introduced, representing the change in monotonicity between the first and second applanation peak of the waveform. The variables mostly related to the PRAA were identified from a set of 41 variables including age, axial length (AL), keratometry, ORA corneal hysteresis, ORA corneal resistant factor, 35 well established ORA waveform parameters, and Monot1-2, using a model selection method based on the second-order bias-corrected Akaike information criterion. The optimal model for PRAA was the AL and six ORA waveform parameters, including Monot1-2. This optimal model was significantly better than the model without Monot1-2 (p = 0.0031, ANOVA). The current study suggested the value of a generative deep learning approach in discovering new useful parameters that may have clinical relevance.
- 20 April 2020
- Published by Springer Nature in Scientific Reports
- Vol. 10  (1) , 1-8
Open AccessVisualizing the dynamic change of Ocular Response Analyzer waveform using Variational Autoencoder in association with the peripapillary retinal arteries angle
The aim of the current study is to identify possible new Ocular Response Analyzer (ORA) waveform parameters related to changes of retinal structure/deformation, as measured by the peripapillary retinal arteries angle (PRAA), using a generative deep learning method of variational autoencoder (VAE). Fifty-four eyes of 52 subjects were enrolled. The PRAA was calculated from fundus photographs and was used to train a VAE model. By analyzing the ORA waveform reconstructed (noise filtered) using VAE, a novel ORA waveform parameter (Monot1-2), was introduced, representing the change in monotonicity between the first and second applanation peak of the waveform. The variables mostly related to the PRAA were identified from a set of 41 variables including age, axial length (AL), keratometry, ORA corneal hysteresis, ORA corneal resistant factor, 35 well established ORA waveform parameters, and Monot1-2, using a model selection method based on the second-order bias-corrected Akaike information criterion. The optimal model for PRAA was the AL and six ORA waveform parameters, including Monot1-2. This optimal model was significantly better than the model without Monot1-2 (p = 0.0031, ANOVA). The current study suggested the value of a generative deep learning approach in discovering new useful parameters that may have clinical relevance.
- 28 February 2020
- Published by BMJ in British Journal of Ophthalmology
- Vol. 104  (12) , 1697-1703
Usefulness of data augmentation for visual field trend analyses in patients with glaucoma
Aim To investigate the usefulness of data augmentation in visual field (VF) trend analyses in patients with glaucoma. Method This study included 6380 VFs from 638 eyes of 417 patients with open-angle glaucoma. Various affine transformations were applied to augment the VF data: (1) rotation, (2) scaling, (3) vertical and horizontal shift and (4) a combination of these different transformations. Using pointwise linear regression (PLR), the total deviation (TD) values of a patient’s 10th VF were predicted using TD values from shorter VF series (from first to third VFs (VF1–3) to first to ninth VFs (VF1–9)) with and without VF data augmentation, and the root mean squared error (RMSE) was calculated. Results With PLR, mean RMSE without VF augmentation averaged from 3.95 (VF1–3) to 19.01 (VF1–9) dB. The RMSE was significantly improved by applying the different transformations: (1) rotation (from VF1–3 to VF1–7), (2) scaling (from VF1–3 to VF1–6), (3) vertical and horizontal shifts (from VF1–3 to VF1–4) and (iv) a combination of these (from VF1–3 to VF1–7). Progression rates in VF1–10 had better agreement with those in shorter VF series when a combination of affine transformation was applied. The differences in rates were between 1.9 (VF1–3) and 0.39 (VF1–9) dB if augmentation was used, which was significantly smaller than that observed when augmentation was not applied (from 2.6 with VF1–3 to 0.26 dB with VF1–9). Conclusion It is useful to apply VF data augmentation techniques when predicting future VF progression in glaucoma using PLR, especially with short VF series.
- 28 February 2020
- Published by BMJ in British Journal of Ophthalmology
- Vol. 104  (12) , 1697-1703
Usefulness of data augmentation for visual field trend analyses in patients with glaucoma
Aim To investigate the usefulness of data augmentation in visual field (VF) trend analyses in patients with glaucoma. Method This study included 6380 VFs from 638 eyes of 417 patients with open-angle glaucoma. Various affine transformations were applied to augment the VF data: (1) rotation, (2) scaling, (3) vertical and horizontal shift and (4) a combination of these different transformations. Using pointwise linear regression (PLR), the total deviation (TD) values of a patient’s 10th VF were predicted using TD values from shorter VF series (from first to third VFs (VF1–3) to first to ninth VFs (VF1–9)) with and without VF data augmentation, and the root mean squared error (RMSE) was calculated. Results With PLR, mean RMSE without VF augmentation averaged from 3.95 (VF1–3) to 19.01 (VF1–9) dB. The RMSE was significantly improved by applying the different transformations: (1) rotation (from VF1–3 to VF1–7), (2) scaling (from VF1–3 to VF1–6), (3) vertical and horizontal shifts (from VF1–3 to VF1–4) and (iv) a combination of these (from VF1–3 to VF1–7). Progression rates in VF1–10 had better agreement with those in shorter VF series when a combination of affine transformation was applied. The differences in rates were between 1.9 (VF1–3) and 0.39 (VF1–9) dB if augmentation was used, which was significantly smaller than that observed when augmentation was not applied (from 2.6 with VF1–3 to 0.26 dB with VF1–9). Conclusion It is useful to apply VF data augmentation techniques when predicting future VF progression in glaucoma using PLR, especially with short VF series.