研究者業績

髙橋 秀徳

タカハシ ヒデノリ  (Hidenori TAKAHASHI)

基本情報

所属
筑波大学 医学医療系 教授
自治医科大学 眼科学講座 准教授 (データサイエンスセンター副長)
学位
博士(医学)(2008年3月 東京大学)

ORCID ID
 https://orcid.org/0000-0001-5331-4730
J-GLOBAL ID
201401070051534987
Researcher ID
H-2945-2019
researchmap会員ID
B000238009

外部リンク

主要な研究キーワード

 6

主要な学歴

 4

主要な論文

 233
  • Satoru Inoda, Hidenori Takahashi, Hitoshi Yamagata, Yoichiro Hisadome, Yusuke Kondo, Hironobu Tampo, Shinichi Sakamoto, Yusaku Katada, Toshihide Kurihara, Hidetoshi Kawashima, Yasuo Yanagi
    Scientific Reports 12(1) 21826-21826 2022年12月17日  査読有り
    Abstract We herein propose a PraNet-based deep-learning model for estimating the size of non-perfusion area (NPA) in pseudo-color fundus photos from an ultra-wide-field (UWF) image. We trained the model with focal loss and weighted binary cross-entropy loss to deal with the class-imbalanced dataset, and optimized hyperparameters in order to minimize validation loss. As expected, the resultant PraNet-based deep-learning model outperformed previously published methods. For verification, we used UWF fundus images with NPA and used Bland–Altman plots to compare estimated NPA with the ground truth in FA, which demonstrated that bias between the eNPA and ground truth was smaller than 10% of the confidence limits zone and that the number of outliers was less than 10% of observed paired images. The accuracy of the model was also tested on an external dataset from another institution, which confirmed the generalization of the model. For validation, we employed a contingency table for ROC analysis to judge the sensitivity and specificity of the estimated-NPA (eNPA). The results demonstrated that the sensitivity and specificity ranged from 83.3–87.0% and 79.3–85.7%, respectively. In conclusion, we developed an AI model capable of estimating NPA size from only an UWF image without angiography using PraNet-based deep learning. This is a potentially useful tool in monitoring eyes with ischemic retinal diseases.
  • Alexandru Lavric, Valentin Popa, Hidenori Takahashi, Rossen M Hazarbassanov, Siamak Yousefi
    Scientific Reports 11(1) 10732-10732 2021年12月  査読有り
    Abstract The main goal of this study is to identify the association between corneal shape, elevation, and thickness parameters and visual field damage using machine learning. A total of 676 eyes from 568 patients from the Jichi Medical University in Japan were included in this study. Corneal topography, pachymetry, and elevation images were obtained using anterior segment optical coherence tomography (OCT) and visual field tests were collected using standard automated perimetry with 24-2 Swedish Interactive Threshold Algorithm. The association between corneal structural parameters and visual field damage was investigated using machine learning and evaluated through tenfold cross-validation of the area under the receiver operating characteristic curves (AUC). The average mean deviation was − 8.0 dB and the average central corneal thickness (CCT) was 513.1 µm. Using ensemble machine learning bagged trees classifiers, we detected visual field abnormality from corneal parameters with an AUC of 0.83. Using a tree-based machine learning classifier, we detected four visual field severity levels from corneal parameters with an AUC of 0.74. Although CCT and corneal hysteresis have long been accepted as predictors of glaucoma development and future visual field loss, corneal shape and elevation parameters may also predict glaucoma-induced visual functional loss.
  • Ichiya Sano, Hidenori Takahashi, Satoru Inoda, Shinichi Sakamoto, Yusuke Arai, Yuji Takahashi, Akira Ohkubo, Hidetoshi Kawashima, Chihiro Mayama
    American journal of ophthalmology 206 11-16 2019年10月  査読有り
    PURPOSE: To investigate changes in the interpupillary distance (IPD) after continual instillation of topical prostaglandin analogs (PGAs) in glaucoma patients as an objective indicator of prostaglandin-associated periorbitopathy (PAP). DESIGN: Retrospective, comparative case series. METHODS: A total of 152 institutional patients with glaucoma were enrolled in this study. Inclusion criteria were visual acuities exceeding 10/20 bilaterally and no intraocular surgery during observation. Intervention/observation procedures: First-time bilateral instillation of bimatoprost, travoprost, latanoprost, or tafluprost and IPDs measured by automatic refractometry. IPDs, intraocular pressures (IOPs), and refractive errors were measured before and after continual drug administration (treatment, 2-24 months). MAIN OUTCOME MEASUREMENTS: Post-treatment changes in IPDs. A total of 61 untreated patients served as controls. RESULTS: The IPDs shortened significantly (P < 0.001) after treatment (-0.80 ± 2.1 mm); the IPDs of control subjects remained unchanged (0.05 ± 0.96 mm; P = 0.69). The IPD change after bimatoprost instillation (-2.20 ± 0.97 mm) was significantly (P < 0.001) greater than with other PGAs (-0.65 ± 2.09 mm). The IOPs decreased significantly (P < 0.001) (-3.7 ± 4.3 mm Hg); the refractive errors did not change significantly (P < 0.099) (-0.07 ± 0.69 diopter) post-treatment. The percentages of subjects with 2-mm or greater decreases in IPD after bimatoprost, travoprost, latanoprost, or tafluprost were 85.7%, 20.0%, 18.2%, and 17.2%, respectively, and with 3-mm or greater decreases in IPD 35.7%, 12.0%, 14.5%, and 12.1%, respectively. The specificities were 93.4% and 100% in the control group, respectively, with IPD threshold changes of 2 and 3 mm or more, respectively. CONCLUSIONS: The IPD decreased significantly after topical PGAs within 24 months. The effect was significantly greater with bimatoprost than with other PGAs. The noninvasive, immediate automatic refractometry measurement may be an objective numerical indicator of PAP.
  • Takahashi H, Inoue Y, Tan X, Inoda S, Sakamoto S, Arai Y, Yanagi Y, Fujino Y, Kawashima H
    Clin Ophthalmol 12 369-375 2018年  査読有り
  • Shinichi Sakamoto, Hidenori Takahashi, Xue Tan, Yuji Inoue, Yoko Nomura, Yusuke Arai, Yujiro Fujino, Hidetoshi Kawashima, Yasuo Yanagi
    British Journal of Ophthalmology 102(4) 448-454 2018年  査読有り
    PURPOSE: To determine changes in multiple cytokine concentrations in the anterior chamber during the induction phase of ranibizumab treatment in patients with neovascular age-related macular degeneration (AMD). METHODS: This prospective study included 48 treatment-naïve neovascular AMD eyes of 48 patients who received three consecutive monthly injections of ranibizumab at the Japan Community Health Care Organization Tokyo Shinjuku Medical Center between November 2010 and August 2012. We collected ~0.2 mL aqueous humour before the first and third (2 months later) injections. Controls were 80 eyes with cataracts without retinal disease. The cytokines C-X-C motif chemokine ligand 1 (CXCL1), interferon-γ-induced protein 10 (IP-10), C-X-C motif chemokine ligand 12 (CXCL12), C-X-C motif chemokine ligand 13 (CXCL13), monocyte chemoattractant protein 1 (MCP-1), CCL11, C-C motif chemokine ligand 11 (CCL11), interleukin-6 (IL-6), interleukin-10 (IL-10) and matrix metalloproteinase 9 (MMP-9) were analysed using multiplex cytokine assays. RESULTS: Mean ages of the patients with AMD and controls were 73 and 75 years, respectively, and 31 (65%) and 37 (46%) subjects were men, respectively. Polypoidal choroidal vasculopathy was found in 27 eyes (56%). Mean concentrations of cytokines in aqueous humour in patients with neovascular AMD before the first and third ranibizumab injections were as follows (in pg/mL): CXCL1, 8.4 and 3.3; IP-10, 110 and 55; CXCL12, 480 and 240; CXCL13, 9.2 and 2.6; MCP-1, 620 and 220; CCL11, 7.1 and 2.8; IL-6, 5.9 and 1.6; IL-10, 0.15 and 0.015 (all p<0.0001), and MMP-9, 0.92 and 1.5 (p=0.0216), respectively. Concentrations of all cytokines decreased significantly after two consecutive ranibizumab injections, except for MMP-9, which increased significantly. CONCLUSIONS: After two monthly consecutive antivascular endothelial growth factor injections, inflammatory cytokine levels in the aqueous humour of the eyes with AMD were strongly suppressed, while MMP-9 levels increased.
  • Hidenori Takahashi
    Plos One 12(6) e0179790-e0179790 2017年6月22日  査読有り
  • Hidenori Takahashi, Yoko Nomura, Junko Nishida, Yujiro Fujino, Yasuo Yanagi, Hidetoshi Kawashima
    INVESTIGATIVE OPHTHALMOLOGY & VISUAL SCIENCE 57(2) 462-466 2016年2月  査読有り
    PURPOSE. Commercially available enzyme-linked immunosorbent assay (ELISA) kits are often used to monitor vascular endothelial growth factor (VEGF) levels in exudative age-related macular degeneration. To test their accuracy, this study performed measurements using the ELISA kits in the presence of anti-VEGF drugs. METHODS. The concentrations of bevacizumab, pegaptanib, or ranibizumab at 28 days and aflibercept at 28 and 56 days after an injection were estimated based on previous pharmacokinetic studies. Vascular endothelial growth factor concentrations were measured with two widely used VEGF ELISA kits in the presence of anti-VEGF drugs or control mouse immunoglobulin G (IgG). The monocyte chemotactic protein-1 (MCP-1) ELISA kit was used as a non-VEGF ELISA control kit. RESULTS. The concentrations of aflibercept, bevacizumab, pegaptanib, and ranibizumab were estimated at 0.14 to 7.2, 4.9, 8.6, and 0.11 to 1.1 mu g/mL, respectively. ELISA underestimated the VEGF concentration 2- to 100-fold lower in the presence of an anti-VEGF drug, except for pegaptanib, at all VEGF concentrations tested (7.8-1500 pg/mL). Vascular endothelial growth factor at 1000 pg/mL was measured as 92, 150, and 170 pg/mL in the presence of aflibercept (7.2 mu g/mL), bevacizumab (4.9 mu g/mL), and ranibizumab (1.1 mu g/mL), respectively (all P &lt; 0.0001), and the measured VEGF concentration decreased proportionately by 90% to 92% with aflibercept, 85% to 94% with bevacizumab, and 83% to 99% with ranibizumab. The control mouse IgG did not interfere with the measurement of VEGF. Ranibizumab did not affect the measurements with MCP-1 ELISA. CONCLUSIONS. Investigators should exercise caution when interpreting measurements of VEGF ELISA in patients being treated with an anti-VEGF drug.
  • Hidenori Takahashi, Yoko Nomura, Xue Tan, Yujiro Fujino, Hidetoshi Kawashima, Yasuo Yanagi
    BRITISH JOURNAL OF OPHTHALMOLOGY 99(8) 1065-1069 2015年8月  査読有り
    Purpose To investigate the association of posterior vitreous detachment (PVD) with aqueous levels of vascular endothelial growth factor (VEGF) and other inflammatory cytokines. Methods These are prospective comparative studies. Subjects comprised 98 eyes for VEGF concentration and 80 eyes for other cytokines, which are normal except for cataract. PVD was examined by B-mode ultrasonography, and the subjects were divided into complete PVD group (PVD group) or the other group (without PVD group). At the beginning of cataract surgery, aqueous humour was collected and the concentrations of VEGF and other inflammatory cytokines were determined using ELISA and a multiplex cytokine assay, respectively. The concentrations of these cytokines were compared between the two groups. Results Complete PVD was observed in 56 (57%) eyes for VEGF concentration analysis, and 51 (64%) eyes for the other cytokines analysis. The concentrations of VEGF, adjusted for the average age, axial length and gender distribution, was 47 pg/ mL in the PVD group and 72 pg/mL in the without PVD group. The concentrations of IP10, MCP-1, CXCL13 and CCL11 were 53, 450, 3.8 and 6.0 pg/ mL in the PVD group, and 100, 560, 7.0 and 8.4 pg/ mL in the without PVD group, respectively. Multiple regression analysis revealed that the logarithmic concentration of VEGF, IP-10, MCP-1, CXCL13 and CCL11 were significantly lower in the eyes with PVD than in those without PVD independently of age, sex and axial length (p = 0.01, p = 0.002, p = 0.009, 0.006 and 0.03, respectively). Conclusions PVD is related to the change in the multiple intraocular inflammatory cytokines.
  • Hidenori Takahashi, Yuko Ohkubo, Aya Sato, Mikiko Takezawa, Yujiro Fujino, Yasuo Yanagi, Hidetoshi Kawashima
    Retina 35(7) 1331-1338 2015年  査読有り
    Background: In age-related macular degeneration, various factors in clinical practice cause delays to arise between the time exudative change is observed and the time antivascular endothelial growth factor drugs are actually injected. We investigated the influence of injection delay on prognosis. Methods: Subjects were 50 eyes (50 patients from 2 hospitals) that were administered ranibizumab monotherapy for age-related macular degeneration for 1 year since exudative change was first observed. We investigated the mean number of delay days for each injection. Results: Mean injection delay was between 0 and 104 days. Significant prognostic factors for visual acuity were initial best-corrected visual acuity (P , 0.01) and mean injection delay (P = 0.03). We estimated that for an initial best-corrected visual acuity of 0.40 logMAR unit (20/50 Snellen equivalent), the respective best-corrected visual acuity values after 1 year for mean injection delays of 0, 7, 14, 28, and 56 days would be 0.22 (20/33), 0.24 (20/35), 0.26 (20/37), 0.31 (20/40), and 0.39 (20/49). For an initial best-corrected visual acuity of 0.097 (20/25), the respective values would be 0.054 (20/23), 0.075 (20/24), 0.10 (20/25), 0.14 (20/28), and 0.22 (20/33). Conclusion: Long-term visual acuity prognosis worsened when scheduling problems delayed intravitreal injection of anti-vascular endothelial growth factor drugs.
  • N Nishiyama, A Iriyama, WD Jang, K Miyata, K Itaka, Y Inoue, H Takahashi, Y Yanagi, Y Tamaki, H Koyama, K Kataoka
    NATURE MATERIALS 4(12) 934-941 2005年12月  査読有り
    The control of gene transfection in the body is a core issue in gene therapy. Photochemical internalization is a technology that allows light-induced delivery of DNA, drugs or other biological factors directly inside cells. Usually it requires that a photosensitizer be added to the drug-delivery system to photochemically destabilize the endosomal membrane. Here we present a system for in vivo DNA delivery in which these two components are assembled into one structure. This is a ternary complex composed of a core containing DNA packaged with cationic peptides and enveloped in the anionic dendrimer phthalocyanine, which provides the photosensitizing action. The ternary complex showed more than 100-fold photochemical enhancement of transgene expression in vitro with reduced photocytotoxicity. In an animal experiment, subconjuctival injection of the ternary complex followed by laser irradiation resulted in transgene expression only in the laser-irradiated site. This work demonstrates a new biomedical application for dendrimers, and the first success in the photochemical-internalization-mediated gene delivery in vivo.

主要なMISC

 77
  • 髙橋秀徳
    臨床眼科 78(2) 190-194 2024年2月15日  招待有り筆頭著者責任著者

主要な書籍等出版物

 2

講演・口頭発表等

 244
  • 髙橋, 秀徳
    SIP第3期「統合型ヘルスケアシステムの構築における生成AIの活用」公開シンポジウム 国立健康危機管理研究機構  招待有り
  • TAKAHASHI, Hidenori
    Conference on Global Health Innovations - Introducing Germany’s Smart Health Hub Bochum in North Rhine-Westphalia (NRW) 筑波大・ボーフム大  招待有り
  • TAKAHASHI, Hidenori
    第129回日本眼科学会総会  招待有り
    The speaker founded DeepEyeVision and has launched a fundus screening support solution, DeepEyeVision for RetinaStation, which displays deviations from normal in fundus photographs, and DeepEyeVision for California, which highlights abnormal retinal vascular density. Features that can be handled by ophthalmologists putting in extra effort had low demand. On the other hand, functions that even specialized ophthalmologists cannot perform were difficult to improve in terms of accuracy. Additionally, when the target disease has a low prevalence, the demand was also low. Diagnostic programs require approval, which costs 100 million yen to obtain. While limiting functionality to fall within the certification scope of existing devices can allow for cheaper certification, recovering development costs remains challenging. Therefore, it is essential to utilize public subsidies to obtain approval or certification, reduce risks and costs, and then present the program to medical device manufacturers. We are currently researching visual field estimation from optical coherence tomography (OCT) images in the high-prevalence area of glaucoma. 【利益相反公表基準】該当有 【倫理審査】承認有 【IC】取得有
  • 高橋, 宏典, 伊野田, 悟, 髙橋, 秀徳, 坂本, 晋一, 長岡, 広祐, 吉田, 花, 橋本, 悠人, 三澤, 真奈美, 蕪城, 俊克, 柳, 靖雄
    第129回日本眼科学会総会
    【目的】私達は未治療新生血管型加齢黄斑性(naïve nAMD)に対する休薬を含めたtreat, extend, and stop-protocol(TES)法によるbrolucizumab休薬率を2年間で47%と報告した。今回、同様のTES法でfaricimabでの休薬率を後ろ向きに調査した。 【方法】自治医科大学附属病院で2022年6月から2023年6月までにfaricimab連続3回の導入期以後、最大投与間隔期間4ヶ月のTAEを行い、4ヶ月毎の間隔で2回dryを確認した場合に休薬としたTES法で治療を受けたnAMD43例43眼。観察期間72週間の休薬率、week16での黄斑部滲出性変化、最高矯正視力(BCVA)/central subfield retinal thickness(CST)/central choroidal thickness(CCT)の変化量、投与回数、安全性、切り替え症例の評価を行った。 【結果】平均年齢74.4歳 (男性27例、62.8%)。治療前平均BCVA、CST、CCT:0.375±0.454、318±84µm、276±75µm。macular neovascularization 28例(type1 14例、type2 6例、type3 8例)、polypoidal choroidal vasculopathy 15例。week16ではwet 4例、dry 31例であり、72週治療継続できた35例および脱落8例中21例(49%)が休薬、BCVA、CST、CCT変化量は0.11±0.39、90±81µm、23±30µm(p=.104、<.001、.001)であった。脱落は8例(眼内炎1例、通院自己中2例、切り替え5例)、15文字以上の視力低下は3例(白内障進行2例、原疾患の増悪1例)であった。 【結論】naïve nAMDに対するfaricimab投与による休薬率は49%であった。 【利益相反公表基準】該当有 【倫理審査】該当無 【IC】取得有
  • 高山, 卓也, 宇都, 翼, 柘植, 泰希, 近藤, 祐亮, 反保, 宏信, 千葉, 真弓, 蕪城, 俊克, 柳, 靖雄, 髙橋, 秀徳
    第129回眼科学会総会
    【目的】網膜裂孔は網膜周辺部に存在することが多く、一般的な眼底画像では検出が難しい。超広角眼底画像は約200度の視野をカバーし網膜の約80%を撮影できるため、網膜裂孔の検出に有効である。しかし広視野が原因で見逃しも多い。本研究は、超広角眼底画像において網膜裂孔を可視化する機械学習モデルを構築し、その性能を評価することである。 【方法】2018年から2021年にかけて自治医科大学で撮影された8083症例、15188眼、35474枚の超広角眼底画像を用いた。裂孔がある435枚の画像と、裂孔がない35009枚の画像に分類し、裂孔がある画像にはピクセルレベルでアノテーションを行った。機械学習モデルにはPraNetを採用し、過学習防止のためのハイパーパラメータの最適化を行った。 【結果】裂孔画像62枚中、102個の裂孔が検出された。感度:0.87、特異度:0.56、陽性的中率:0.22、陰性的中率:0.97であった。裂孔に対する結果は、感度:78.4%、陽性的中率:56.7%、Recall: 0.685±0.378、Precision: 0.357±0.235、セントロイド距離スコア:0.113±0.242、Diceスコア:0.435±0.258であった。偽陽性の原因は、網膜出血:22.4%、レーザー痕:20.7%、画像範囲外:20.1%、睫毛:4.0%であった。 【結論】超広角眼底画像において網膜裂孔を検出するPraNetを用いた機械学習モデルを構築した。出血やレーザー痕、画像範囲外を偽陽性とすることが多く、さらなる改善が必要である。 【利益相反公表基準】該当有 【倫理審査】承認有 【IC】該当無
  • 伊野田, 悟, 髙橋, 秀徳, 坂本, 晋一, 長岡, 広祐, 吉田, 花, 橋本, 悠人, 高橋, 宏典, 三澤, 真奈美, 蕪城, 俊克, 柳, 靖雄
    第129回眼科学会総会
    【目的】私達はbrolucizumabに起因した眼内炎症(IOI)では、非IOI群と比較して血中のCD4+(Th)細胞のうち、Th2細胞の割合が有意に少ないことを報告した。今回、faricimab(IVF)治療後IOIを発症した新生血管加齢黄斑変性(nAMD)3症例においてTh細胞の割合について評価を行った。 【対象と方法】自治医科大学附属病院において、2022年6月から2024年9月までにnAMDに対するIVFを開始し、血液検査に同意を得たnAMD患者109人109眼中3人3眼でIOIが発症した。flow cytometryでIL-4とIFN-γの陽性細胞と陰性細胞を検出し、Th1、Th2に分類し、正常群(平均年齢75.2(95%CI 73.3-77.0、男性76例))とその割合を比較した。 【結果】3例とも導入期を含みIVF加療を開始。症例1:75歳男性、左Macular neovascularization (MNV) type3において、初回・最終投与より137日・4日後に強膜炎・虹彩毛様体炎を発症。症例2:70歳男性、右PCVにおいて、初回・最終投与より224日・98日後に虹彩毛様体炎を発症。症例3:80女性、右MNV type1において、初回・最終投与より119日・58日後に虹彩毛様体炎・硝子体混濁を発症。正常群におけるTh1/Th2は11.7 (7.83 to 15.5) であり、IOI群では8.42、5.80、11.0とTh2が低い傾向が認められた。 【結論】nAMD患者でのIVFに起因したIOIでは血中Th2細胞数が少ない傾向があり、brolucizumabに起因するIOIと同様の機序が関連している可能性が示唆された。 【利益相反公表基準】該当有 【倫理審査】承認有 【IC】取得有
  • 高山, 卓也, 宇都, 翼, 柘植, 泰希, 近藤, 佑亮, 反保, 宏信, 千葉, 真弓, 蕪城, 俊克, 柳, 靖雄, 髙橋, 秀徳
    第85回栃木県眼科集談会
  • TAKAHASHI, Hidenori, Ichimiya, Shunsuke, Tampo, Hironobu, Tsuge, Taiki, Noda, Ryotaro, Kondo, Yusuke, Senoo, Yuki, Hirono, Kazushi, Inoda, Satoru, Kaburaki, Toshikatsu, Yanagi, Yasuo, Kadonosono, Kazuaki
    FUJI RETINA 日本網膜硝子体学会  招待有り
    Purpose: Preoperative understanding of the ciliary body length allows for optimal placement of the scleral incision, improving instrument access. We previously reported a method for measuring ciliary body length using anterior segment OCT images and machine learning in eyes with long axial lengths. However, in eyes with normal axial lengths, thicker sclera results in lower-quality OCT images. In this study, we used eyes with both long and normal axial lengths. Methods: A total of 261 anterior segment OCT images from 190 eyes of 111 patients were obtained from two institutions. The accuracy of the predicted coordinates was assessed by calculating the root mean square error (RMSE) and mean absolute error (MAE) relative to the ground truth coordinates. Coordinate estimation was conducted using three approaches: UNet-based segmentation, template matching, and a hybrid method combining both techniques. Results: Coordinate estimation using UNet segmentation resulted in an MAE of 1.77 mm and an RMSE of 1.61. The application of template matching led to improved accuracy, reducing the MAE to 0.586 and the RMSE to 1.06. Further enhancement was achieved by integrating both methods, yielding an MAE of 0.329 and an RMSE of 0.357. No significant correlation was found between axial length and estimation accuracy. Conclusions: We developed a method combining UNet-based segmentation and template matching to estimate the pars plana position from anterior segment OCT images with an accuracy of less than 0.5 mm. This level of precision may contribute to safer insertion site determination for vitreoretinal surgery and intravitreal injections.
  • 反保, 宏信, Kushagra, Shekhawa, 宇都, 翼, 柘植, 泰希, 近藤, 佑亮, 伊野田, 悟, 柳, 靖雄, 髙橋, 秀徳
    第5回日本眼科AI学会総会 秋山雅人
    【目的】 深層学習による網膜裂孔領域検出では、眼底写真全体に占める網膜裂孔領域が小さく不均衡なため、検出モデルが全画像に対して網膜裂孔なしと予測するように収束してしまう場合がある。そこで、学習時の画像切り抜きによるデータ拡張に着目し、画像に占める網膜裂孔領域を増やすことで、検出モデルの学習安定と精度向上を図った。 【対象と方法】 自治医科大学附属病院、たかはし眼科、斎藤眼科、とつか眼科のOptos California(Nikon, Tokyo)で撮影された超広角眼底写真(UWF)から、網膜裂孔を有する806枚と有しない37724枚をランダムに選択し、深層学習を行った。 画像に占める網膜裂孔領域の割合を増やすため、網膜裂孔を有する画像を特定のサイズで切り抜き、必ず網膜裂孔領域を含める前処理を行った(実験群)。対照群では、画像のランダムな箇所を切り抜く前処理を行った。 【結果】 網膜裂孔を有する73枚のUWFでの評価を行った。対照群のDice Score 0.496±0.340、 Recall 0.597±0.357、Precision 0.499±0.355であるのに対して、実験群のDice Score 0.604± 0.315、Recall 0.668±0.321、Precision 0.620±0.332 であった。 【結論】 網膜裂孔を選択的に切り抜く前処理の追加によって、深層学習モデルの精度が有意に改善した。
  • 滝波, 俊平, 森川, 翔平, 髙橋, 秀徳, 大鹿, 哲郎
    第5回日本眼科AI学会総会 秋山雅人
  • 髙橋, 秀徳
    第78回日本臨床眼科学会  招待有り
    【目的】超広角走査型レーザー検眼鏡の擬似カラー画像(UWF™)から網膜無灌流領域(NPA)を推定する深層学習モデルを、4施設の多施設データで構築したので報告する。 【対象と方法】4施設のOptos® California(Nikon, Tokyo)で撮影されたUWF(虚血性網膜疾患2,654例4,327眼13,076撮影15,597枚と非虚血性網膜疾患5,112例9,019眼14,859撮影18,391枚からNPAを有する823枚および有しない9725枚をランダムに選択)と超広角蛍光眼底造影画像(UWF-FA)をアノテーションして得られたNPAを用いてPraNetで深層学習した。Bland-Altman プロットを用い、141例150眼154枚のUWF(うちNPA+は54枚)に対して、実測NPAに対する推定NPAの面積を比較した。 【結果】面積Bland-Altmanプロットでは、固定誤差である推定NPAと実測NPAの面積差平均(79.2 mm2, P < 0.0001)は、誤差の許容範囲(462 mm2)より小さかった。外れ値となったUWF像は5.8%で認められた。比例誤差は有意であった(P < 0.0001)。 【結論】4施設から構築したNPA推定アルゴリズムは、誤差は許容範囲に収まったが、比例誤差が有意であった。
  • 土屋, 匠麿, 中川, 怜旺奈, 反保, 宏信, 柘植, 泰希, 間山, 千尋, 近藤, 佑亮, 髙橋, 秀徳, 伊野田, 悟, 蕪城, 俊克, 宮下, 洋, 柳, 靖雄
    第78回日本臨床眼科学会
    【目的】 本研究では眼底画像から緑内障のNFLD領域を自動的に検出するAIモデルを開発し、その性能を評価する。 【対象と方法】 自治医科大で収集した通常画角カラー眼底画像から、緑内障と診断の付いた画像を収集した。収集した画像を緑内障専門医が再度読影し、NFLDのラベルを付与した。NFLDのラベルが付与された画像から、視能訓練士がNFLD領域を描画し、学習・評価用のデータとして採用した。 NFLD陽性の眼底画像データのうち、1451枚(211患者、298眼)を学習用データとした。211患者の平均年齢は57.2歳であった。NFLD陽性の画像データセットに対して、NFLD陰性画像の枚数を1対1になるようにランダムに抽出して、NFLD陰性の学習用データとした。PraNetのネットワークアーキテクチャを採用し、眼底画像の学習を行った。 【結果】 NFLD陽性のテスト画像185枚のうち、183枚が陽性と判定され、領域の検出が行われた。感度は98.9%であり、各画像のDice Scoreの平均は57.8%だった。NFLD陰性のテスト画像185枚のうち、陰性と判定された画像は114枚で、特異度は61.6%だった。 【結論】 眼底画像を用いた学習により、NFLD領域を高い確率で自動検出できるAIモデルの実現が示唆された。今後、各パラメータの調整を行うことで、偽陽性の減少やより正確なNFLD領域の検出が可能なAIモデルの実現が期待される。
  • 伊野田, 悟, 柘植, 泰希, 反保, 宏信, 近藤, 佑亮, 髙橋, 秀徳, 蕪城, 俊克, 柳, 靖雄
    第78回日本臨床眼科学会
    【目的】 本研究では眼底画像からハンフリー視野計の出力を予測するAIモデルを開発し、その性能を評価する。 【対象と方法】 自治医科大で収集したハンフリー視野計のデータと前後1年以内に撮影した通常画角眼底カラー画像を1:1のペアとして学習用データとして使用する。ハンフリー視野計のデータは中心24-2の閾値マップを採用し、固視不良率、偽陽性率、偽陰性率のいずれかが1/3以上のデータは除外した。ハンフリー視野計の学習用データは871例(501患者、729眼)、患者の平均年齢は58.4歳であった。 眼底画像を入力として、HFAの閾値マップ、MD値、PSD値をそれぞれ自動計算する機械学習モデルを作成した。学習にはResNet18のネットワークアーキテクチャを採用した。 【結果】 評価用眼底画像からHFAの閾値マップ、MD値、PSD値の予測を行った。 各画像に対する閾値マップのRMSEの平均値は6.82dBであった。 MDの実測値の平均は-7.52±8.87dBに対してAIによる予測値は-5.16±0.47dBであった。実測値と予測値の誤差の平均は6.48±6.37dBであった。 PSDの実測値の平均が5.51±4.36dBに対してAIによる予測値は4.93±0.72dBであった。PSDの実測値と予測値の誤差の平均は3.85±2.34dBであった。 【結論】 眼底画像を用いた学習により、視野計の出力を自動計算するAIモデルの実現が示唆された。
  • 杉山, 丈実, 伊野田, 悟, 髙橋, 秀徳, 吉田, 花, 高橋, 宏典, 三澤, 真奈美, 高山, 卓也, 松本, 大蔵, 土, 屋 匠麿, 蕪城, 俊克
    第78回日本臨床眼科学会
  • TAKAHASHI, Hidenori
    The 17th Joint Meeting of Japan-Korea-China Ophthalmologists  招待有り
    In 2012, the advent of deep learning technology allowed artificial intelligence (AI) to surpass human capabilities in image recognition by 2015. The world’s first deep learning AI to receive medical approval was the AI for diabetic retinopathy diagnosis in the U.S. in 2018. Since then, a small number of diagnostic AIs have been approved in developed countries, including AIs for noise reduction in imaging, estimation of optic disc cupping volume from fundus photographs, and heatmap-based visualization of abnormal regions on color fundus images. Meanwhile, AIs capable of diagnosing multiple diseases from fundus photographs are available online globally for public use. For example, TeleMedC’s EyeQX Kiosk operates in places such as subway stations in Singapore, where it performs AI-based, non-mydriatic fundus photo diagnoses on the spot. The company also offers RAPIDDx, a service available for free trial online. In the U.S., the non-profit organization Cybersight provides free online access to fundus diagnosis AI. My venture, DeepEyeVision, founded out of Jichi Medical University, has also developed a diagnostic support AI for remote screening of fundus images, covering nearly 100 diseases. In addition, we have obtained certification for AI that visualizes abnormal regions on fundus photographs using heatmaps, and AI that identifies regions of abnormal retinal vascular density from ultra-widefield color fundus images. This presentation provides an overview of the mechanisms behind these AIs. Furthermore, recent advances include AI capable of performing segmentation and diagnosis from optical coherence tomography (OCT) images, many of which have received approval. 2001 Graduated from the University of Tokyo, Faculty of Medicine 2008 Completed graduate studies at the University of Tokyo 2012 Lecturer, Department of Ophthalmology, Jichi Medical University 2016 Founded DeepEyeVision 2017 Associate Professor, Department of Ophthalmology, Jichi Medical University 2020 Deputy Director, Data Science Center, Jichi Medical University 2024 Professor, Center for Cyber Medicine Research, University of Tsukuba
  • 髙橋, 秀徳
    第28回北区医師会医学会 北区医師会  招待有り
  • 髙橋, 秀徳
    北関東 眼疾患 セミナー 2024  招待有り
  • 髙橋, 秀徳
    AMD診療エキスパートフォーラム  招待有り
  • 髙橋, 秀徳
    眼科AI Expert Meeting ノバルティスファーマ株式会社  招待有り
  • TAKAHASHI, Hidenori
    Tsukuba Global Science Week 2024 筑波大学  招待有り
  • 髙橋, 秀徳
    眼科経営学会 眼科経営学会  招待有り
  • 髙橋, 秀徳
    アイリーア8mg新発売記念講演会 in 関東甲信越 vol.1  招待有り
  • TAKAHASHI, Hidenori
    EURETINA 2024 EURETINA
    Purpose: To elucidate the subtypes and prognosis of dry Age-related Macular Degeneration (dry AMD) among Japanese patients defined as geographic atrophy size over 250 µm diameter. To calculate the incidence and progression rates of the various subtypes of dry AMD. Setting/Venue: Forty consecutive cases of dry AMD patients attending Jichi Medical University were enrolled. Methods: The patients had longitudinal data derived from Fundus Autofluorescence (FAF) imaging using a laser scanning ophthalmoscope over intervals exceeding three months. For bilateral cases, the eye with the larger area of atrophy fitting within the image frame was selected. Annotations for complete retinal pigment epithelium and outer retinal atrophy (cRORA) were performed using FAF by three independent annotators. The data were analyzed using Analysis of Variance (ANOVA) and t-tests. Results: A total of 40 eyes with dry AMD (average age of 74 years (SD 8.2) and 50% males) had baseline complete retinal pigment epithelium and outer retinal atrophy (cRORA) area of 9.5 mm² (13), which increased to 15 mm² (18) at the final observation after an average follow-up of 2.3 years, progressing at a rate of 1.6 mm²/year (2.4). Subgroup analysis revealed: Focal type was in 10 eyes and patients had an average age of 69 years, with males constituting 50% of this subgroup. The initial cRORA area was 4.9 mm², expanding to 6.5 mm² after an average of 1.6 years, with a progression rate of 0.25 mm²/year. Banded type: 8 eyes, 78 years, 50%, 18 mm², 2 years, and 0.36 mm²/year. Diffuse type: 10 eyes, 77 years, 30%, 7.1 mm², 2.9 years, and 3.7 mm²/year. Diffuse trickling type: 4 eyes, 70 years, 50%, 7.3 mm², 1.3 years, and 3.0 mm²/year. Patchy type: 4 eyes, 78 years, 100%, 2.9 mm2, 1.0 years, and 0.86 mm2/year. There was no significant difference in age between groups. However, a significant difference was observed in the progression rate, with the diffuse type demonstrating a faster progression compared to the pachy, banded, and focal types (P < 0.05). Conclusions: This study identified the FAF pattern classification in the Japanese population. Compared to Caucasian population reported by FAM-Study Group, there were fewer diffuse and more focal types.
  • Takayama, Takuya, Uto, Tsubasa, Tsuge, Taiki, Kondo, Yusuke, Tanpo, Hironobu, TAKAHASHI, Hidenori, Chiba, Mayumi, Yanagi, Yasuo
    EURETINA 2024
    Purpose: Retinal holes, often located in the retina's peripheral regions, are difficult to detect with standard fundus imaging. Ultra-widefield fundus imaging overcomes these limitations, enabling the identification of these holes. However, the extensive field of view and the small size of the holes mean that detection can be time-consuming and prone to oversights. Therefore, this study aims to leverage AI analysis with a machine learning model to enhance the speed of detection and minimize missed retinal holes through visualization. Setting/Venue: The study utilized ultra-widefield fundus images from 8083 patients at the Jichi Medical University, comprising a total of 15188 eyes and 35474 images, as data for training and evaluating the machine learning model. Methods: Ultra-widefield fundus images were initially prepared for the creation of the dataset, and then classified into two categories, those with retinal holes and those without. There were 435 and 35009 images respectively. Images with retinal holes were annotated at the pixel level to indicate the hole locations on the images. Finally, all prepared images were split into three datasets: train, val and test. In the training dataset, there were 62 images with holes and 34001 images without holes. Similarly, the validation dataset included 53 images with holes and 601 images without holes, while the test dataset consisted of 62 images with holes and 437 images without holes. After that, supervised learning was conducted with a PraNet-based architecture model, which was originally for the accurate polyp segmentation in colonoscopy images, to train on retinal hole detection. In the training phase, for the purpose of improving accuracy, hyperparameter optimization was conducted. To evaluate the pixel-wise detection, we adopt Dice, Recall, Precision and centroid distance as metrics. Specifically, we employed Dice score as the main metric for the hyperparameter tuning, which is the ratio of the intersection size of predicted region and actual hole region to the mean size of the both regions. Results: After the training phase, the model was evaluated using the prepared test dataset. The evaluation metrics yielded the following results: Dice score was 0.435 ± 0.258, Recall was 0.685 ± 0.378, Precision was 0.357 ± 0.235, and Centroid Distance was 0.113 ± 0.242. The low Centroid Distance score indicates the model's success in accurately locating the holes, albeit with a lower than expected Dice score, suggesting difficulties in precisely defining the boundaries between holes and their backgrounds. The Recall and Precision scores further imply that the model tended to overestimate the size of the actual holes. Conclusion: The evaluation results underscore the challenges in achieving precise boundary detection of retinal holes using the current model. Despite effective localization of holes, as indicated by the Centroid Distance, the model's lower Dice score reflects its limitations in accurately delineating the edges of the holes. The disparity between Recall and Precision suggests an inclination towards overestimation of hole areas. Future work will focus on improving boundary precision, potentially by adopting the Intersection over Union (IoU) metric for a stricter evaluation of boundary overlap, aiming to enhance the model's accuracy in identifying retinal holes.
  • TAKAHASHI, Hidenori, Tsuge, Taiki, Chiba, Mayumi, Esmaeelpour, Marieh, Kondo, Yusuke, Inoda, Satoru, Yamaguchi, Chingning Taffeta, Tanpo, Hironobu, Hashimoto, Yuto, Kimura, Kyouta, Kaburaki, Toshikatsu, Yanagi, Yasuo
    EURETINA 2024 EURETINA
    Purpose: To elucidate the subtypes and prognosis of dry Age-related Macular Degeneration (dry AMD) among Japanese patients defined as geographic atrophy size over 250 µm diameter. To calculate the incidence and progression rates of the various subtypes of dry AMD. Setting/Venue: Forty consecutive cases of dry AMD patients attending Jichi Medical University were enrolled. Methods: The patients had longitudinal data derived from Fundus Autofluorescence (FAF) imaging using a laser scanning ophthalmoscope over intervals exceeding three months. For bilateral cases, the eye with the larger area of atrophy fitting within the image frame was selected. Annotations for complete retinal pigment epithelium and outer retinal atrophy (cRORA) were performed using FAF by three independent annotators. The data were analyzed using Analysis of Variance (ANOVA) and t-tests. Results: A total of 40 eyes with dry AMD (average age of 74 years (SD 8.2) and 50% males) had baseline complete retinal pigment epithelium and outer retinal atrophy (cRORA) area of 9.5 mm² (13), which increased to 15 mm² (18) at the final observation after an average follow-up of 2.3 years, progressing at a rate of 1.6 mm²/year (2.4). Subgroup analysis revealed: Focal type was in 10 eyes and patients had an average age of 69 years, with males constituting 50% of this subgroup. The initial cRORA area was 4.9 mm², expanding to 6.5 mm² after an average of 1.6 years, with a progression rate of 0.25 mm²/year. Banded type: 8 eyes, 78 years, 50%, 18 mm², 2 years, and 0.36 mm²/year. Diffuse type: 10 eyes, 77 years, 30%, 7.1 mm², 2.9 years, and 3.7 mm²/year. Diffuse trickling type: 4 eyes, 70 years, 50%, 7.3 mm², 1.3 years, and 3.0 mm²/year. Patchy type: 4 eyes, 78 years, 100%, 2.9 mm2, 1.0 years, and 0.86 mm2/year. There was no significant difference in age between groups. However, a significant difference was observed in the progression rate, with the diffuse type demonstrating a faster progression compared to the pachy, banded, and focal types (P < 0.05). Conclusions: This study identified the FAF pattern classification in the Japanese population. Compared to Caucasian population reported by FAM-Study Group, there were fewer diffuse and more focal types.
  • 髙橋, 秀徳
    DME Expert Meeting in Fukuoka ノバルティスファーマ株式会社  招待有り
  • 髙橋, 秀徳
    東大眼科茶話会 東大眼科同窓会
  • 髙橋, 秀徳
    サイバーメディスン研究センターキックオフ サイバーメディスン研究センター  招待有り
  • 髙橋, 秀徳
    山梨県眼疾患フォーラム2024 千寿製薬株式会社  招待有り
    滲出型加齢黄斑変性(AMD)は慢性疾患であり、長期予後を考慮して治療を進めたい。ラニビズマブの投与タイミング遅れによる予後変化を定量した研究、黄斑をドライにし過ぎることによる視力低下検出の研究を紹介する。また演者はAIの画像識別能がヒトを超えた2015年からAIの研究を始めた。パソコンの組み立てから始め、自治医大糖尿病センターにあった1方向及び4方向のカラー眼底写真それぞれの糖尿病網膜症病期分類1万セットを用いた、1方向からの4方向病期推測を行い、初めての特許出願を契機に起業した。自治医科大学発ベンチャー認定1号を取得し、ベンチャーキャピタルからの出資と指導から様々なAIプログラムを開発した。眼底健診読影の支援AIプログラム、健常逸脱可視化と血管密度逸脱可視化の認証AI医療機器プログラム、これから社会実装をしていきたい超広角眼底写真における網膜裂孔や剥離の検出AIプログラム、OCTからの視野推測プログラムを紹介する。
  • 髙橋, 秀徳
    SIPテーマ2:Med-LMMキックオフミーティング 原田 達也  招待有り
  • 伊野田悟, 髙橋秀徳
    第39回JSCRS学術総会 2024年6月28日  招待有り
  • 髙橋, 秀徳, 田淵, 仁志
    第39回JSCRS学術総会 日本白内障屈折矯正手術学会
  • 髙橋秀徳
    第10回九州DMEフォーラム 2024年6月29日  招待有り
  • 髙橋, 秀徳
    第10回九州DMEフォーラム バイエル薬品株式会社・参天製薬株式会社  招待有り
  • 髙橋, 秀徳
    第39回JSCRS学術総会 日本白内障屈折矯正手術学会  招待有り
    障手術後、糖尿病患者において眼内炎症サイトカインが上昇することが知られている。また、重症の糖尿病網膜症患者は血糖コントロールが不良である可能性が高く、白内障も進行しやすい。したがって、白内障手術を計画する際には、糖尿病患者の眼底状態の把握は重要である。白内障手術の適切な時期を選択し、より頻繁に眼底検査を行うことは、糖尿病網膜症の発生と増悪を予防するために有用である。その一方で、白内障手術を検討する段階では眼底透見性が悪く、検眼鏡的な糖尿病網膜症の病期診断は過小評価となることがある。蛍光眼底造影は、眼底透見性が悪い患者において病期を診断するのに有用である。特に、超広角走査型レーザー検眼鏡登場以後、一度に眼底の80%以上を撮影できるようになり、中心部のみならず周辺網膜の変化を含め広角で観察できるようになった。ただし、その侵襲性から頻回に行うことは難しい。私達は深層学習を用い、カラー眼底写真からNPA領域の推定を行った。10視神経乳頭サイズ、30乳頭サイズのNPA基準に検討では、感度・特異度ともに80%前後という良好な結果が得られた。(Inoda S, et al. Sci Rep 2022) 本講演では、追加データ学習を行った結果と、白内障の本研究結果への影響を発表したい。
  • 髙橋, 秀徳
    第10回九州DMEフォーラム バイエル薬品株式会社・参天製薬株式会社  招待有り
  • 髙橋, 秀徳
    糖尿病黄斑浮腫アップデート〜明日から役立つ治療のポイント〜 ノバルティスファーマ株式会社
  • 髙橋, 秀徳
    RVO適応拡大記念講演会 in Yakushiji 中外製薬株式会社  招待有り
  • 髙橋秀徳
    第2回 Okinawa Retina Summit 2024年5月11日  招待有り
  • Hidenori Takahashi, Takufumi Sakura, Hiroshi Miyashita, Yusuke Kondo, Satoru Inoda, Takuya Takayama, Rossen M. Hazarbassanov, Siamak Yousefi, Kosuke Nagaoka, Hidetoshi Kawashima, Yasuo Yanagi
    ARVO2024 2024年5月9日
  • Rossen M. Hazarbassanov, Ali H. AI-Timemy, Carlos Guilermo Arce, Pedro Henrique Fabres Franco, Luzia Alves Dos Santos, Zahraa Mosa, Hazem Abdelmotaal, Alexandru Lavric, Hidenori Takahashi, Suphi Taneri, Wuqaas Wuqaas, Siamak Yousefi
    ARVO2024 2024年5月6日
  • Hazarbassanov, Rossen M., Al-Zubaidi, Laith, Al-Timemy, Ali, Arce, Carlos, Fabres Franco, Pedro Henrique, Alves Dos Santos, Luzia, Mosa, Zahraa, Abdelmotaal,Hazem M., Lavric, Alexandru, Takahashi, Hidenori, Taneri, Suphi, Munir, Wuqaas, Yousefi, Siamak
    ARVO Imaging in the Eye Conference 2024年5月4日
  • Hidenori Takahashi, Taiki Tsuge, Yusuke Kondo, Satoru Inoda, Takuya Takayama, Rossn M. Hazarbassanov, Siamak Yousefi, Kosuke Nagaoka, Hidetoshi Kawashima, Yasuo Yanagi
    ARVO Imaging in the Eye Conference 2024年5月4日
  • 髙橋 秀徳
    第128回日本眼科学会総会 2024年4月20日
  • 伊野田 悟, 髙橋秀徳, 佐藤 彩, 高橋良太, 橋本悠人, 吉田 花, 高橋宏典, 三澤真奈美, 柳靖雄, 川島秀俊
    第128回日本眼科学会総会 2024年4月19日
  • 吉田 花, 伊野田悟, 髙橋秀徳, 高橋良太, 橋本悠人, 高橋宏典, 三澤真奈美, 柳靖雄, 川島秀俊
    第128回日本眼科学会総会 2024年4月19日
  • 松本大蔵, 宇都翼, 伊野田悟, 近藤祐亮, 劉弘毅, 東洋介, 髙橋宏典, 反保宏信, 橋本悠人, 長岡広祐, 髙橋秀徳, 川島秀俊, 柳靖雄
    第128回日本眼科学会総会 2024年4月18日
  • 松本大蔵, 宇都翼, 伊野田悟, 近藤佑亮, 劉弘毅, 東洋介, 高橋宏典, 高山卓也, 反保宏信, 橋本悠人, 長岡広祐, 髙橋秀徳, 川島秀俊, 柳靖雄
    第83回栃木県眼科集談会 2024年4月14日
  • 髙橋秀徳
    豊島区眼科医会学術講演会 2024年3月30日  招待有り

主要な共同研究・競争的資金等の研究課題

 5

主要な学術貢献活動

 10

主要な社会貢献活動

 1