[論文レビュー] State-of-the-Art Periorbital Distance Prediction and Disease Classification Using Periorbital Features

George R. Nahass, Sasha Hubschman+10 more|arXiv (Cornell University)|Sep 27, 2024

Medical Imaging and Analysis被引用数 5

ひとこと要約

本論文は、最先端の精度を達成する periorbital segmentation パイプラインを開発し、ID および OOD 設定の下で病気分類の特徴として periorbital distance の予測を検討し、OOD タスクで CNNs を上回る。

ABSTRACT

Periorbital distances are critical markers for diagnosing and monitoring a range of oculoplastic and craniofacial conditions. Manual measurement, however, is subjective and prone to intergrader variability. Automated methods have been developed but remain limited by standardized imaging requirements, small datasets, and a narrow focus on individual measurements. We developed a segmentation pipeline trained on a domain-specific dataset of healthy eyes and compared its performance against the Segment Anything Model (SAM) and the prior benchmark, PeriorbitAI. Segmentation accuracy was evaluated across multiple disease classes and imaging conditions. We further investigated the use of predicted periorbital distances as features for disease classification under in-distribution (ID) and out-of-distribution (OOD) settings, comparing shallow classifiers, CNNs, and fusion models. Our segmentation model achieved state-of-the-art accuracy across all datasets, with error rates within intergrader variability and superior performance relative to SAM and PeriorbitAI. In classification tasks, models trained on periorbital distances matched CNN performance on ID data (77--78\% accuracy) and substantially outperformed CNNs under OOD conditions (63--68\% accuracy vs. 14\%). Fusion models achieved the highest ID accuracy (80\%) but were sensitive to degraded CNN features under OOD shifts. Segmentation-derived periorbital distances provide robust, explainable features for disease classification and generalize better under domain shift than CNN image classifiers. These results establish a new benchmark for periorbital distance prediction and highlight the potential of anatomy-based AI pipelines for real-world deployment in oculoplastic and craniofacial care.

研究の動機と目的

• Motivate periorbital distance as a robust biomarker for oculoplastic and craniofacial conditions.
• Develop a domain-specific segmentation pipeline for healthy eyes and compare with existing models.
• Evaluate how predicted periorbital distances can serve as features for disease classification under ID and OOD conditions.

提案手法

• Train a domain-specific segmentation model on a healthy-eye dataset.
• Compare segmentation performance against Segment Anything Model (SAM) and PeriorbitAI.
• Extract periorbital distances from segmentation outputs and use as features for classification with shallow, CNN, and fusion models.
• Evaluate in-distribution (ID) and out-of-distribution (OOD) performance.
• Assess explainability and robustness of anatomy-based features versus CNN features.

実験結果

リサーチクエスチョン

• RQ1 Can a domain-specific segmentation model achieve state-of-the-art periorbital segmentation accuracy across diverse disease classes and imaging conditions?
• RQ2 Do periorbital distance features enable competitive disease classification under ID conditions compared to CNNs?
• RQ3 Are distance-based features more robust than CNN features under domain shifts (OOD)?
• RQ4 What is the impact of fusion models on ID performance and their sensitivity under OOD shifts?

主な発見

• Segmentation model achieves state-of-the-art accuracy with error within intergrader variability and outperforms SAM and PeriorbitAI.
• Distance-based features yield 77–78% accuracy on ID disease classification, matching CNN performance.
• Under OOD conditions, distance-based methods outperform CNNs by a large margin (63–68% vs. 14%).
• Fusion models reach the highest ID accuracy (≈80%) but degrade when CNN features are weakened under OOD shifts.
• Overall, segmentation-derived distances offer robust and explainable features that generalize better under domain shift than CNN image classifiers.