[論文レビュー] Hierarchical 3D fully convolutional networks for multi-organ segmentation

Holger R. Roth, Hirohisa Oda+6 more|arXiv (Cornell University)|Apr 21, 2017

Advanced Neural Network Applications参考文献 12被引用数 119

ひとこと要約

本論文は、CTスキャンにおける多臓器セグメンテーションのための二段階の粗→細い3D FCN（3D U-Netに基づく）を提案し、臓器ごとに平均Diceを7.5ポイント改善、未知データでの強力な性能を実現します。

ABSTRACT

Recent advances in 3D fully convolutional networks (FCN) have made it feasible to produce dense voxel-wise predictions of full volumetric images. In this work, we show that a multi-class 3D FCN trained on manually labeled CT scans of seven abdominal structures (artery, vein, liver, spleen, stomach, gallbladder, and pancreas) can achieve competitive segmentation results, while avoiding the need for handcrafting features or training organ-specific models. To this end, we propose a two-stage, coarse-to-fine approach that trains an FCN model to roughly delineate the organs of interest in the first stage (seeing $\sim$40% of the voxels within a simple, automatically generated binary mask of the patient's body). We then use these predictions of the first-stage FCN to define a candidate region that will be used to train a second FCN. This step reduces the number of voxels the FCN has to classify to $\sim$10% while maintaining a recall high of $>$99%. This second-stage FCN can now focus on more detailed segmentation of the organs. We respectively utilize training and validation sets consisting of 281 and 50 clinical CT images. Our hierarchical approach provides an improved Dice score of 7.5 percentage points per organ on average in our validation set. We furthermore test our models on a completely unseen data collection acquired at a different hospital that includes 150 CT scans with three anatomical labels (liver, spleen, and pancreas). In such challenging organs as the pancreas, our hierarchical approach improves the mean Dice score from 68.5 to 82.2%, achieving the highest reported average score on this dataset.

研究の動機と目的

• 臓器特異モデルを用いずにCTでの正確な多臓器セグメンテーションを動機づける。
• 階層的で粗→細のFCNが難しい境界に焦点を合わせられることを示す。
• 異なる病院のデータセット間での一般化を示す。
• 腹部の7構造における Dice スコアの改善を定量化する。

提案手法

• 3D U-Net FCNを用いて八クラスをセグメントする（七つの臓器＋背景）。
• 訓練は二段階で行う：Stage 1は体尺度の候補領域を概略的に定義する（ボクセルの約40%程度）。
• Stage 2はより小さな候補領域内でセグメンテーションを精査する（ボクセルの約10%程度）。
• 入力をダウンサンプリングして視野を広げ、推論時にスライディングタイル戦略を適用。
• 前景と背景のボクセルをバランスさせるため、ボクセルワイズ交差エントロピー損失をクラス重み付きで適用。
• データ拡張としてランダムな3Dエラスティック変形とランダム回転で訓練する。

実験結果

リサーチクエスチョン

• RQ1 Can a single 3D FCN model segment multiple abdominal organs without organ-specific models?
• RQ2 Does a hierarchical coarse-to-fine approach improve segmentation accuracy, especially for small/thin structures?
• RQ3 How well does the method generalize to unseen data from a different hospital and scanner?

主な発見

• Mean Dice score improvement of 7.5 percentage points per organ in validation (Stage 2 vs Stage 1).
• Stage 1 recall > 99% with about 10% false positives when dilating candidate regions by r=3.
• Arteries improved from 59.0 to 79.6 Dice in Stage 2; pancreas improved from 54.8 to 63.1 Dice in Stage 2.
• On unseen testing data (liver, spleen, pancreas), Stage 2 non-overlapping achieved mean Dice of 68.5 for pancreas, 93.2 for liver, and 89.7 for spleen; overlapping tiling improved pancreas to 82.2 Dice.
• The method achieved state-of-the-art-like results on an unseen dataset, with liver and pancreas performance benefiting from the hierarchical approach.