{"data":{"featured":{"edges":[{"node":{"frontmatter":{"title":"Double-Mix Pseudo-Label Framework","cover":{"childImageSharp":{"gatsbyImageData":{"layout":"constrained","placeholder":{"fallback":"data:image/webp;base64,UklGRswAAABXRUJQVlA4IMAAAABwBQCdASoUABQAPtFgqE+oJSOiKAgBABoJZwDOdaYboOm7dYtVkWT5qvPkgc3yt32jB8jgAP7n4xI8OcQPL7vy8YKyUk3QDIrv7gEKJCMIVXlisRaINnZkw5ne38Ba3zY5afGCtotObGH4TDxZryNQQOsz8wuuoCB9Am37W0oDUr1+gBxbtQhA1Gu+2Y5uREvX1r6kJCZZuXpjeYVf7nvx0VjBqRHnrTvgGNsbaW+FRQ19Zh90ajfFQMKb37LQAAA="},"images":{"fallback":{"src":"/static/a51b4988d697ddbe2b9aaddf74f2f431/17e5b/cover.webp","srcSet":"/static/a51b4988d697ddbe2b9aaddf74f2f431/d9d38/cover.webp 175w,\n/static/a51b4988d697ddbe2b9aaddf74f2f431/335a6/cover.webp 350w,\n/static/a51b4988d697ddbe2b9aaddf74f2f431/17e5b/cover.webp 700w,\n/static/a51b4988d697ddbe2b9aaddf74f2f431/eedfd/cover.webp 1400w","sizes":"(min-width: 700px) 700px, 100vw"},"sources":[{"srcSet":"/static/a51b4988d697ddbe2b9aaddf74f2f431/6c74f/cover.avif 175w,\n/static/a51b4988d697ddbe2b9aaddf74f2f431/c90d5/cover.avif 350w,\n/static/a51b4988d697ddbe2b9aaddf74f2f431/91a97/cover.avif 700w,\n/static/a51b4988d697ddbe2b9aaddf74f2f431/f5a86/cover.avif 1400w","type":"image/avif","sizes":"(min-width: 700px) 700px, 100vw"}]},"width":700,"height":695}}},"tech":["PyTorch","MONAI","Semi-supervised Learning","CT Segmentation"],"github":"https://github.com/lzhang30/Double-Mix","external":"https://link.springer.com/article/10.1007/s11548-024-03281-1","cta":""},"html":"

少数アノテーションのCTボリュームにおけるカテゴリ不均衡問題を解決する半教師あり学習フレームワーク。Double-Mix疑似ラベル手法により、ラベルなしデータから高品質な監督信号を生成し、臓器・病変セグメンテーション精度を大幅に向上。腹部マルチ臓器・肝臓腫瘍・膵臓などの公開データセットで最先端性能を達成し、International Journal of Computer Assisted Radiology and Surgery（2025）に採択・掲載。

"}},{"node":{"frontmatter":{"title":"In-Body CA — 体型・内臓形状推定AI","cover":{"childImageSharp":{"gatsbyImageData":{"layout":"constrained","placeholder":{"fallback":"data:image/jpeg;base64,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"},"images":{"fallback":{"src":"/static/abde78b14404a846dd623cc50c239435/9398b/cover.jpg","srcSet":"/static/abde78b14404a846dd623cc50c239435/8c618/cover.jpg 175w,\n/static/abde78b14404a846dd623cc50c239435/eb152/cover.jpg 350w,\n/static/abde78b14404a846dd623cc50c239435/9398b/cover.jpg 700w","sizes":"(min-width: 700px) 700px, 100vw"},"sources":[{"srcSet":"/static/abde78b14404a846dd623cc50c239435/50167/cover.avif 175w,\n/static/abde78b14404a846dd623cc50c239435/5a8a9/cover.avif 350w,\n/static/abde78b14404a846dd623cc50c239435/bc4f6/cover.avif 700w","type":"image/avif","sizes":"(min-width: 700px) 700px, 100vw"},{"srcSet":"/static/abde78b14404a846dd623cc50c239435/81b0f/cover.webp 175w,\n/static/abde78b14404a846dd623cc50c239435/bfc62/cover.webp 350w,\n/static/abde78b14404a846dd623cc50c239435/74bac/cover.webp 700w","type":"image/webp","sizes":"(min-width: 700px) 700px, 100vw"}]},"width":700,"height":467}}},"tech":["Python","PyTorch","3D Slicer","Unity","Open3D"],"github":"","external":"https://in-body-ca.org/about.html","cta":""},"html":"

単一正面写真から骨格・体型を推定し、110万例規模のCTデータベース由来の臓器モデルを個人の体型に連動させる可視化システム。日本内閣府ムーンショット目標1の研究成果として2025年大阪・関西万博（EXPO 2025）にて展示。デモ要件定義・アルゴリズム開発・チーム調整・進捗管理を全て担当。

📰 横尾太郎氏インタビュー記事（JST Moonshot） ｜ 🔗 万博展示 JSTムーンショット公式ポスト

"}},{"node":{"frontmatter":{"title":"CT読影支援VLM — 所見自動生成AI","cover":{"childImageSharp":{"gatsbyImageData":{"layout":"constrained","placeholder":{"fallback":"data:image/jpeg;base64,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"},"images":{"fallback":{"src":"/static/ac7dda01743fc40c488f67f151b7d827/2b365/cover.jpg","srcSet":"/static/ac7dda01743fc40c488f67f151b7d827/1849b/cover.jpg 175w,\n/static/ac7dda01743fc40c488f67f151b7d827/3f5de/cover.jpg 350w,\n/static/ac7dda01743fc40c488f67f151b7d827/2b365/cover.jpg 700w","sizes":"(min-width: 700px) 700px, 100vw"},"sources":[{"srcSet":"/static/ac7dda01743fc40c488f67f151b7d827/dae43/cover.avif 175w,\n/static/ac7dda01743fc40c488f67f151b7d827/d7667/cover.avif 350w,\n/static/ac7dda01743fc40c488f67f151b7d827/7ec1a/cover.avif 700w","type":"image/avif","sizes":"(min-width: 700px) 700px, 100vw"},{"srcSet":"/static/ac7dda01743fc40c488f67f151b7d827/5d873/cover.webp 175w,\n/static/ac7dda01743fc40c488f67f151b7d827/26a00/cover.webp 350w,\n/static/ac7dda01743fc40c488f67f151b7d827/f23f0/cover.webp 700w","type":"image/webp","sizes":"(min-width: 700px) 700px, 100vw"}]},"width":700,"height":394}}},"tech":["PyTorch","Hugging Face","LoRA / PEFT","CT画像解析","VLM"],"github":"","external":"https://www.nagoya-u.ac.jp/researchinfo/result/2025/04/20250424_i2.html","cta":""},"html":"

CT検査の差分を比較し、放射線科医の読影を支援する日本語所見レポート自動生成VLM。数十万例のCT画像から有用症例を自動選別し、CT画像特徴と日本語テキストを統合。独自改良したMoELoRA手法を適用し高品質なレポート生成を実現。研究成果は名古屋大学公式サイトに掲載され、SIP第3期公開シンポジウム（2025年4月）にて発表。論文誌へ投稿中。

"}},{"node":{"frontmatter":{"title":"Top Competition Results","cover":{"childImageSharp":{"gatsbyImageData":{"layout":"constrained","placeholder":{"fallback":"data:image/jpeg;base64,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"},"images":{"fallback":{"src":"/static/2efa3753dc0208814b4443a5e9471765/c00de/cover.jpg","srcSet":"/static/2efa3753dc0208814b4443a5e9471765/a9efa/cover.jpg 175w,\n/static/2efa3753dc0208814b4443a5e9471765/4bdfd/cover.jpg 350w,\n/static/2efa3753dc0208814b4443a5e9471765/c00de/cover.jpg 700w","sizes":"(min-width: 700px) 700px, 100vw"},"sources":[{"srcSet":"/static/2efa3753dc0208814b4443a5e9471765/30dfb/cover.avif 175w,\n/static/2efa3753dc0208814b4443a5e9471765/f30ea/cover.avif 350w,\n/static/2efa3753dc0208814b4443a5e9471765/a1761/cover.avif 700w","type":"image/avif","sizes":"(min-width: 700px) 700px, 100vw"},{"srcSet":"/static/2efa3753dc0208814b4443a5e9471765/21ab0/cover.webp 175w,\n/static/2efa3753dc0208814b4443a5e9471765/adbff/cover.webp 350w,\n/static/2efa3753dc0208814b4443a5e9471765/69a36/cover.webp 700w","type":"image/webp","sizes":"(min-width: 700px) 700px, 100vw"}]},"width":700,"height":328}}},"tech":["Python","PyTorch","Deep Learning","NLP / 音声解析"],"github":"","external":"https://www.kaggle.com/ngyzly","cta":""},"html":"

Kaggle Competitions Expert（全体 Rank 550 / 207,133）。Silver ×5・Bronze ×2 のメダルを保有。多分野にわたり Top 1–2% の成績を達成：

🥈 NVIDIA Nemotron Model Reasoning Challenge：Rank 65 / 4,182（Top 1.5%）
🥈 Kaggle Deep Past — アッカド語→英語翻訳：Rank 22 / 3,148（Top 0.6%）
🥈 Kaggle BirdCLEF+ 2026 — 鳥類音声識別：Rank 34 / 4,092（Top 0.8%）
🥈 Kaggle Vesuvius Challenge — 古文書解析：Rank 20 / 1,391（Top 1.4%）

"}}]}}}