Dual-input convolutional neural network for glaucoma diagnosis using spectral-domain optical coherence tomography.

BACKGROUND/AIMS: To evaluate, with spectral-domain optical coherence tomography (SD-OCT), the glaucoma-diagnostic ability of a deep-learning classifier. METHODS: A total of 777 Cirrus high-definition SD-OCT image sets of the retinal nerve fibre layer (RNFL) and ganglion cell-inner plexiform layer (GCIPL) of 315 normal subjects, 219 patients with early-stage primary open-angle glaucoma (POAG) and 243 patients with moderate-to-severe-stage POAG were aggregated. The image sets were divided into a training data set (252 normal, 174 early POAG and 195 moderate-to-severe POAG) and a test data set (63 normal, 45 early POAG and 48 moderate-to-severe POAG). The visual geometry group (VGG16)-based dual-input convolutional neural network (DICNN) was adopted for the glaucoma diagnoses. Unlike other networks, the DICNN structure takes two images (both RNFL and GCIPL) as inputs. The glaucoma-diagnostic ability was computed according to both accuracy and area under the receiver operating characteristic curve (AUC). RESULTS: For the test data set, DICNN could distinguish between patients with glaucoma and normal subjects accurately (accuracy=92.793%, AUC=0.957 (95% CI 0.943 to 0.966), sensitivity=0.896 (95% CI 0.896 to 0.917), specificity=0.952 (95% CI 0.921 to 0.952)). For distinguishing between patients with early-stage glaucoma and normal subjects, DICNN's diagnostic ability (accuracy=85.185%, AUC=0.869 (95% CI 0.825 to 0.879), sensitivity=0.921 (95% CI 0.813 to 0.905), specificity=0.756 (95% CI 0.610 to 0.790)]) was higher than convolutional neural network algorithms that trained with RNFL or GCIPL separately. CONCLUSION: The deep-learning algorithm using SD-OCT can distinguish normal subjects not only from established patients with glaucoma but also from patients with early-stage glaucoma. The deep-learning model with DICNN, as trained by both RNFL and GCIPL thickness map data, showed a high diagnostic ability for discriminatingpatients with early-stage glaucoma from normal subjects.

Comparison of changes of macular ganglion cell-inner plexiform layer defect between stable group and progression group in primary open-angle glaucoma.

PURPOSE: To compare the changes in macular ganglion cell-inner plexiform layer (GCIPL) defect between stable and progression primary open-angle glaucoma (POAG) groups. STUDY DESIGN: Retrospective observational study. METHODS: A total of 100 POAG eyes with localized retinal nerve fiber layer (RNFL) defect and corresponding macular GCIPL defect were selected for this study. Glaucoma progression was defined by either structural or functional deterioration. The number of abnormal superpixels on macular GCIPL deviation maps was calculated using a customized MATLAB program. GCIPL defect change was evaluated in two aspects: increased angular width and increased area. The defect patterns were categorized and compared between the stable and progression groups. RESULTS: The increase rate of angular width of GCIPL defect was higher in the progression group than in the stable group (P = 0.029). In respect to the area of GCIPL defect, there was no statistically significant differences between the groups (P = 0.227). Twenty-seven (27) of 100 (27.0%) eyes showed increased angular width of GCIPL defect. It was more frequent in the progression group than in the stable group (P = 0.043). Seventeen (17) of 27 (63.0%) eyes showed the away from the horizontal temporal raphe type progression and it was the most common change pattern of angular width of GCIPL defect. CONCLUSIONS: Increased angular width of GCIPL defect was the more prominent feature of change, and was more frequent in the progression group than in the stable group. Among the types of GCIPL defect classified, the away from the horizontal temporal raphe type was the most common.

White blood cell differential count of maturation stages in bone marrow smear using dual-stage convolutional neural networks.

The white blood cell differential count of the bone marrow provides information concerning the distribution of immature and mature cells within maturation stages. The results of such examinations are important for the diagnosis of various diseases and for follow-up care after chemotherapy. However, manual, labor-intensive methods to determine the differential count lead to inter- and intra-variations among the results obtained by hematologists. Therefore, an automated system to conduct the white blood cell differential count is highly desirable, but several difficulties hinder progress. There are variations in the white blood cells of each maturation stage, small inter-class differences within each stage, and variations in images because of the different acquisition and staining processes. Moreover, a large number of classes need to be classified for bone marrow smear analysis, and the high density of touching cells in bone marrow smears renders difficult the segmentation of single cells, which is crucial to traditional image processing and machine learning. Few studies have attempted to discriminate bone marrow cells, and even these have either discriminated only a few classes or yielded insufficient performance. In this study, we propose an automated white blood cell differential counting system from bone marrow smear images using a dual-stage convolutional neural network (CNN). A total of 2,174 patch images were collected for training and testing. The dual-stage CNN classified images into 10 classes of the myeloid and erythroid maturation series, and achieved an accuracy of 97.06%, a precision of 97.13%, a recall of 97.06%, and an F-1 score of 97.1%. The proposed method not only showed high classification performance, but also successfully classified raw images without single cell segmentation and manual feature extraction by implementing CNN. Moreover, it demonstrated rotation and location invariance. These results highlight the promise of the proposed method as an automated white blood cell differential count system.

Automated network analysis to measure brain effective connectivity estimated from EEG data of patients with alcoholism.

OBJECTIVE: Detection and diagnosis based on extracting features and classification using electroencephalography (EEG) signals are being studied vigorously. A network analysis of time series EEG signal data is one of many techniques that could help study brain functions. In this study, we analyze EEG to diagnose alcoholism. APPROACH: We propose a novel methodology to estimate the differences in the status of the brain based on EEG data of normal subjects and data from alcoholics by computing many parameters stemming from effective network using Granger causality. MAIN RESULTS: Among many parameters, only ten parameters were chosen as final candidates. By the combination of ten graph-based parameters, our results demonstrate predictable differences between alcoholics and normal subjects. A support vector machine classifier with best performance had 90% accuracy with sensitivity of 95.3%, and specificity of 82.4% for differentiating between the two groups.

Patterns of glaucoma progression in retinal nerve fiber and macular ganglion cell-inner plexiform layer in spectral-domain optical coherence tomography.

PURPOSE: To evaluate the progressive changes of circumpapillary retinal nerve fiber layer (RNFL) and macular ganglion cell-inner plexiform layer (GCIPL) thicknesses measured by spectral-domain optical coherence tomography (Cirrus SD-OCT) in open-angle glaucoma. METHODS: One hundred-fourteen eyes of open-angle glaucoma patients with localized RNFL defect who had 3 years' worth of annual RNFL photography and OCT measurements were enrolled in this retrospective study. The progression rates of serial RNFL and GCIPL thicknesses (µm), angular width (°) and area (mm2) of defect on RNFL and GCIPL deviation maps were determined by linear mixed-effect models. RESULTS: Over a mean follow-up period of 3.16 years, 50 patients out of a total of 114 patients were classified as progressors based on the structural evaluation. The progressors showed significantly higher progression rates in average, 6 and 11 o'clock sector RNFL thicknesses, angular width and area of defect in RNFL deviation map, as well as inferotemporal and minimum GCIPL thicknesses than the non-progressors. The thicknesses of the 6 o'clock sector RNFL and minimum GCIPL exhibited the highest reduction rates among the RNFL and GCIPL parameters assessed, respectively. CONCLUSIONS: When evaluating glaucoma progression by OCT, careful observation of the average, 6 and 11 o'clock sectors in RNFL and inferotemporal and minimum GCIPL thicknesses can be helpful. We can effectively assess early changes of glaucoma progression with GCIPL thickness as well as RNFL thickness.

Glaucoma-Diagnostic Ability of Ganglion Cell-Inner Plexiform Layer Thickness Difference Across Temporal Raphe in Highly Myopic Eyes.

PURPOSE: To evaluate the glaucoma-diagnostic ability of the ganglion cell-inner plexiform layer (GCIPL) thickness difference across the temporal raphe in highly myopic eyes. METHODS: We consecutively enrolled a total of 195 highly myopic eyes (axial length [AL] >26.5 mm) of 195 subjects: 93 glaucoma patients along with and 102 nonglaucomatous subjects. Cirrus high-definition optical coherence tomography (OCT) was employed to scan all of the subjects' macular and optic discs. Using a MATLAB-based customized program (the GCIPL hemifield test), a positive test result was automatically declared if the following two conditions were met: (1) the horizontal line is detected for longer than one-half of the distance from the temporal inner elliptical annulus to the outer elliptical annulus, and (2) the average GCIPL thickness difference within 10 pixels of the reference line, both above and below, is 5 µm or more. The glaucoma-diagnostic ability was computed using the area under the receiver operating characteristic curve (AUC). RESULTS: Among the glaucomatous eyes, GCIPL hemifield test positivity was shown in 92.5% (86 of 93), significantly higher than that for the nonglaucomatous eyes (4.90%, 5 of 102; P <0.001). The value of AUC for the GCIPL hemifield test was excellent (0.938; sensitivity 92.50%, specificity 95.10%) and was the best compared with those for any of OCT parameters. CONCLUSIONS: In highly myopic eyes, determination of the presence or absence of GCIPL thickness difference across the temporal raphe via OCT macula scan can be a useful means of distinguishing the glaucomatous damage.

Influence of the Rotational Alignment of the Femoral and Patellar Components on Patellar Tilt in Total Knee Arthroplasty.

PURPOSE: The purpose of this study was to analyze the influence of rotational alignment of the femoral and patellar components on patellar tilt after total knee arthroplasty (TKA). MATERIALS AND METHODS: A total of 56 patients (76 knees) who underwent TKA using Advance Medial Pivot Knee system between May 2009 and April 2011 and were available for minimum 1-year follow-up were enrolled in this study. Whiteside's line and the transepicondylar line were used to determine the rotational alignment of the femoral component. Patella cut was aimed to be parallel to the anterior patellar cortex during surgery. Radiographic evaluation was performed using plain axial radiographs. The rotational alignment of the femoral component was measured as the angle between the anterior condylar axis and the surgical transepicondylar axis. The patellar resection angle was measured between the patellar resection axis and the anterior cortical line of the patella. Patellar tilt was evaluated to investigate the correlation with the rotation of the femoral component and patellar resection angle. RESULTS: The mean rotation of the femoral component was 0.42°±3.18° of internal rotation. The mean patellar resection angle was 1.82°±3.44°, indicating medial overresection. The mean patellar tilt was 6.12°±4.31° of lateral tilt. The rotational angle of the femoral component showed a negative correlation with patellar tilt in the linear regression analysis (p=0.749), but the patellar resection angle showed a positive correlation with patellar tilt (p<0.001). CONCLUSIONS: Accurate patellar resection is recommended for proper patellar tracking in TKA.

Automated Detection of Hemifield Difference across Horizontal Raphe on Ganglion Cell--Inner Plexiform Layer Thickness Map.

PURPOSE: A MATLAB-based (The MathWorks, Inc, Natick, MA) computer program (the ganglion cell-inner plexiform layer [GCIPL] hemifield test) for automated detection of GCIPL thickness difference across the horizontal raphe was developed, and its glaucoma diagnostic performance was assessed. DESIGN: Cross-sectional study. PARTICIPANTS: A total of 65 eyes of normal, healthy subjects along with 162 eyes of patients with glaucoma (79 preperimetric and 83 early perimetric). METHODS: Cirrus high-definition optical coherence tomography (HD-OCT) (Carl Zeiss Meditec, Dublin, CA) was used to scan all of the subjects' macular and optic discs. A positive (i.e., "outside normal limits") GCIPL hemifield test result was declared if the following 3 conditions were all met: (1) The reference line (a horizontal line dividing the superior and inferior hemifields) is continuously detected for longer than one-half of the distance from the temporal inner elliptical annulus to the outer elliptical annulus; (2) the average GCIPL thickness difference within 10 pixels of the reference line, both above and below, is ≥5 µm; and (3) the average RGB color ranges of the 10 pixels above and below the reference line display blue in 1 hemifield and red/yellow/white in the other hemifield. MAIN OUTCOME MEASURES: Comparison of diagnostic ability using the areas under the receiver operating characteristic curves (AUCs). RESULTS: A positive GCIPL hemifield test result was observed more frequently in the glaucomatous eyes (74/79 preperimetric, 78/83 early perimetric) than in the normal eyes (1/65). In the preperimetric group, the AUC of the GCIPL hemifield test (0.967; sensitivity 94.94%, specificity 98.46%) was greater than that of the minimum GCIPL thickness (0.933), the inferotemporal GCIPL thickness (0.907), and the average GCIPL thickness (0.899) (P=0.09, 0.06, and 0.03, respectively). In the early perimetric group, the AUC of the GCIPL hemifield test (0.962; sensitivity 93.98%, specificity 96.46%) was greater than that of the inferotemporal GCIPL thickness (0.938), the minimum GCIPL thickness (0.919), and the average GCIPL thickness (0.912) (P=0.38, 0.17, and 0.11, respectively). CONCLUSIONS: For discrimination of early glaucomatous structural loss, most notably in preperimetric glaucoma cases, identification of the GCIPL thickness difference across the horizontal raphe was effective.

Long-Term Reproducibility of Macular Ganglion Cell Analysis in Clinically Stable Glaucoma Patients.

PURPOSE: To investigate the long-term reproducibility of macular ganglion cell analysis in clinically stable glaucoma patients using spectral-domain optical coherence tomography (SD-OCT). METHODS: One hundred nine eyes of 109 clinically stable open-angle glaucoma patients with a localized retinal nerve fiber layer (RNFL) defect and a corresponding macular ganglion cell-inner plexiform layer (GCIPL) defect were included in this retrospective, longitudinal study. Clinical stability was defined as showing no change on serial structural (stereo-disc and RNFL photography) and functional (visual field progression analysis) assessments. Three serial SD-OCT (Cirrus-HD) peripapillary and macular scans taken at 6-month intervals were analyzed. Intraclass correlation coefficient (ICC), coefficient of variation (CV), test-retest standard deviation (TRTSD), and tolerance limit of area and angular width of GCIPL defect and GCIPL thickness measurements were assessed. RESULTS: The ICC of the GCIPL thickness parameters ranged from 0.966 to 0.992, and the CV from 2.0% to 5.5%. The TRTSD was the lowest for the average GCIPL thickness (1.45 µm), the highest for the minimum GCIPL thickness (3.42 µm), and varied from 1.54 to 2.16 µm for the sectoral measurements. The ICC, CV, and TRTSD were 0.993, 3.9%, and 5.32° for angular width, and 0.930, 6.7%, and 0.27 mm2 for area of GCIPL defect. Measurement variances (TRTSD) for the GCIPL measurements showed no significant association with the glaucomatous severity. CONCLUSIONS: The macular GCIPL thickness and deviation maps showed excellent long-term intervisit reproducibility. Macular ganglion cell analysis can be considered as an effective means of monitoring glaucomatous progression in macula.

Correlation between depth and area of retinal nerve fiber layer defect as measured by spectral domain optical coherence tomography.

BACKGROUND: To evaluate the correlation between the depth and area of retinal nerve fiber layer (RNFL) defect, as measured on an RNFL map of spectral-domain optical coherence tomography (SD-OCT). METHODS: The RNFL of 472 glaucoma subjects and of 217 healthy subjects was imaged by an SD-OCT. RNFL defect depth and area on the RNFL map were expressed as an RNFL defect depth percentage index (RDPI) and an RNFL defect area index (RDAI), respectively, according to the following two formulas: 100×[1-{summation of thicknesses of RNFL defects/summation of thicknesses of upper 95th percentile range of age-matched healthy subjects in areas corresponding to defects}]; 100×[number of superpixels of RNFL defects/(46 × 46-superpixels inside optic disc or ß zone parapapillary atrophy)]. The best-fitting model describing the relationship between the two parameters was derived by fractional polynomial analysis. RESULTS: Logarithmic fit was determined to be the best-fitting model in describing the relationship of the RDPI against the RDAI (y = 53.4 + 3.7 ln(x) and y = 50.9 + 5.9 ln(x) in superior and inferior hemifields, respectively). The expected RDAIs at the point where the RDPI and RDAI rates of change were the same were 3.7 and 5.9 %; the corresponding upper 95 % confidence interval limits of the RDPI 59.0 and 61.8 % in the superior and inferior hemifields, respectively. CONCLUSIONS: The correlation between the RNFL defect depth and area, as derived from the RNFL map, was best described by the logarithmic fit. Changes were more marked in depth than in area, especially for mild localized defects.

LV peak instantaneous wall stress versus time-stress-integral as measures of afterload in aortic stenosis.

BACKGROUND: LV pressure overload triggers biological events leading to LV hypertrophy, and the severity of LV pressure overload has been represented by LV wall stress (LVWS). However, duration of exposure to the given LVWS has not been considered as a factor leading to LV hypertrophy. METHODS: In 20 patients with severe aortic stenosis, LV pressure, with a Millar catheter, and LV dimensions, with echocardiography, were simultaneously obtained. By synchronising LV pressure waveform and LV dimensions, the instantaneous peak LVWS (IP-LVWS) and systolic LV wall time-stress-integral (LV TSI), which is the LVWS integrated during the systolic period and physically implies LV wall impulse per unit area, were obtained. These measurements were obtained at rest, during right atrial pacing at an HR of 10% higher than resting HR, and after esmolol infusion. RESULTS: IP-LVWS cannot be predicted from the peak LV pressure or peak dp/dt. There were no significant differences in peak IP-LVWS at rest, during pacing or after esmolol infusion. However, systolic LV TSI after esmolol infusion was significantly higher compared to that at rest or during pacing. Only LV TSI correlated with LV wall thickness or LV mass at rest (r=0.64, p=0.002 and r=0.54, p=0.014, respectively), while correlations between IP-LVWS and LV wall thickness or LV mass at rest were not statistically significant (r=0.32, p=0.173 and r=0.32, p=0.168, respectively). CONCLUSIONS: As IP-LVWS cannot be predicted from the peak LV pressure or peak dp/dt, one should be cautious in interpreting the data regarding peak LV pressure as the parameter representing LV afterload. Based on the current results, the systolic LV TSI might be more important than IP-LVWS in LV hypertrophy. In addition, as LV TSI is affected by the systolic duration, factors prolonging systolic duration might exert adverse effects on LV hypertrophy.

Reproducibility of spectral-domain optical coherence tomography RNFL map for glaucomatous and fellow normal eyes in unilateral glaucoma.

PURPOSE: To compare the reproducibility of the optical coherence tomography (OCT) retinal nerve fiber layer (RNFL) thickness map between glaucomatous and fellow normal eyes of unilateral glaucoma patients. METHODS: In this prospective case-control study, Cirrus HD-OCT was performed for 79 unilateral glaucoma patients 3 times on the first visit and on 3 subsequent visits within a 2-month period. Test-retest standard deviation (TRT-SD) and tolerance limit based on the 1.645×√2×TRT-SD formula were derived for RNFL thicknesses at the respective superpixels of the RNFL thickness map. RESULTS: The TRT-SDs and tolerance limits of the glaucomatous eyes (TRT-SD: 2.75 to 20.25 µm; tolerance limits: 6.40 and 47.11 µm) were significantly smaller than those of the fellow normal eyes (TRT-SD: 2.73 to 26.49 µm; tolerance limits: 6.35 and 61.63 µm) in the superotemporal, inferotemporal, and superonasal areas (P <0.05). The TRT-SDs in most areas showed a significant positive correlation with the RNFL thicknesses (P<0.05). CONCLUSIONS: The test-retest variabilities of the Cirrus HD-OCT RNFL thickness map of the glaucomatous eyes were lower than those of the fellow normal eyes, especially in areas of high diagnostic importance. Moreover, variability was positively correlated with the baseline RNFL thicknesses. Therefore, adjusting the tolerance limits on the basis of the baseline RNFL thickness values might help improve the ability to recognize progression. Further prospective studies on this issue are warranted.

Intramolecular ionic Diels-Alder reactions of alpha-acetylenic acetals.

The intramolecular ionic Diels-Alder reaction of alpha-acetylenic acetals as a precursor of the propargyl cation has been investigated in the presence of Lewis acids and in protic acids. The reaction of diene-tethered alpha-acetylenic acetals (1-2) with formic acid yielded the regioselective intramolecular ionic Diels-Alder reaction products, bicyclodienal (9) and bicyclodienone (11) derivatives, in good yields.