Learn Fine-grained Adaptive Loss for Multiple Anatomical Landmark Detection in Medical Images

Guang-Quan Zhou, Juzheng Miao, Xin Yang, Rui Li, En-Ze Huo, Wenlong Shi, Yuhao Huang, Jikuan Qian, Chaoyu Chen, Dong NiarXiv 2021 · DOI · arXiv

TL;DR

This paper introduces LaOML (Landmark-Aware Objective Metric Learning), a framework that uses reinforcement learning to automatically tune the target precision—specifically the Gaussian standard deviation of ground-truth heatmaps—for medical landmark detection. By optimizing these hyperparameters dynamically for each landmark, the model achieves higher localization accuracy and better training stability than traditional methods that rely on hand-crafted heuristics.

What Problem Does This Paper Address?

Detecting anatomical landmarks (like the crown or rump in a fetal ultrasound) is a cornerstone of medical image analysis. The state-of-the-art approach is heatmap regression, where a neural network learns to produce a "likelihood map" for each landmark.

The "ground truth" for these heatmaps is typically a Gaussian distribution centered on the landmark. However, researchers face a frustrating trade-off when setting the standard deviation ($\sigma$) of this Gaussian:

  1. Small $\sigma$: Provides high precision but creates a "needle in a haystack" problem during training because the target signal is extremely sparse.
  2. Large $\sigma$: Makes training easier and more stable but results in blurry predictions and lower final accuracy.

Most existing works use a fixed, global $\sigma$ for all landmarks. This ignores the fact that different anatomical structures have varying levels of visual ambiguity and uncertainty. Manually tuning $\sigma$ for dozens of different landmarks is cumbersome and task-specific.

Methods

The authors propose a "learning-to-learn" framework that optimizes the network weights and the target precision simultaneously.

Schematic view of the LaOML framework, showing the U-Net backbone, the RL controllers, and the early-stop strategy.
  • RL-based Controller: Instead of a human setting $\sigma$, the framework assigns a Reinforcement Learning (RL) agent (a small MLP) to each landmark. The agent's "state" is the current training progress, its "action" is to increase, decrease, or maintain the current $\sigma$, and its "reward" is the localization accuracy on a validation set.
  • Bilevel Optimization: The system alternates between two phases. In the inner loop, the main detection network (a U-Net) is trained using the current $\sigma$ values. In the outer loop, the RL agents update their policies to find $\sigma$ values that maximize validation accuracy.
  • Early-Stop Strategy: To prevent the RL agents from oscillating or over-exploring once a good precision is found, the authors introduced an early-stop mechanism. If the variance of the distance error for a specific landmark drops below a threshold, the search for that landmark's $\sigma$ is terminated, locking in the optimal value.

Datasets

The authors evaluated their method on two distinct medical imaging tasks:

  1. In-house Prenatal Ultrasound (US) Dataset: 511 fetal midsagittal plane images with 8 landmarks (e.g., thalamus, nasal bone, rump).
  2. Public Cephalometric X-Ray Dataset (ACXRLDC): 400 lateral cephalograms from the ISBI 2015 challenge, featuring 19 landmarks.

Key Insights and Ideas

  • Hyperparameters as Learnable States: The core "aha moment" is treating the loss function's parameters as dynamic variables that should adapt to the model's current capability.
  • Landmark-Specific Precision: The study confirms that different landmarks "prefer" different $\sigma$ values. As shown in the dynamic curves, some landmarks converge to a very sharp precision ($\sigma \approx 3$), while others require a wider net ($\sigma > 15$) to account for anatomical ambiguity.
  • Decoupling Optimization: By using RL, the authors decouple the optimization of the objective metric from the optimization of the network weights. This avoids the biases found in other adaptive loss methods (like Adaloss) that rely solely on the rate of loss change.

Key Results

LaOML consistently outperformed baseline U-Nets and other refinement strategies like GANs or RNNs.

  • Accuracy: On the ultrasound dataset, LaOML reduced the Mean Radial Error (MRE) from 2.22mm to 1.80mm. On the X-ray dataset, it achieved an MRE of 1.39mm, outperforming several state-of-the-art methods.
  • Robustness: The method proved particularly effective at reducing "extreme errors"—cases where standard models completely miss a landmark due to noise or poor contrast.
Visual comparison showing how LaOML (right) produces more accurate landmark predictions compared to the baseline and other refinement methods.

Limitations and Future Work

While effective, the authors noted a few areas for improvement:

  • Update Granularity: The current action space for the RL agent is coarse (simple additive updates of $\pm 1$). A more continuous or fine-grained update strategy might yield even better results.
  • Initial Value Sensitivity: Although the RL agent explores the space, the final performance still has a slight dependency on the initial $\sigma$ value.
  • Loss Function Design: The authors suggest that future work could involve designing entirely new loss functions that are inherently more sensitive to small $\sigma$ values, further pushing the boundaries of precision.

Summary updated

Abstract

Automatic and accurate detection of anatomical landmarks is an essential operation in medical image analysis with a multitude of applications. Recent deep learning methods have improved results by directly encoding the appearance of the captured anatomy with the likelihood maps (i.e., heatmaps). However, most current solutions overlook another essence of heatmap regression, the objective metric for regressing target heatmaps and rely on hand-crafted heuristics to set the target precision, thus being usually cumbersome and task-specific. In this paper, we propose a novel learning-to-learn framework for landmark detection to optimize the neural network and the target precision simultaneously. The pivot of this work is to leverage the reinforcement learning (RL) framework to search objective metrics for regressing multiple heatmaps dynamically during the training process, thus avoiding setting problem-specific target precision. We also introduce an early-stop strategy for active termination of the RL agent's interaction that adapts the optimal precision for separate targets considering exploration-exploitation tradeoffs. This approach shows better stability in training and improved localization accuracy in inference. Extensive experimental results on two different applications of landmark localization: 1) our in-house prenatal ultrasound (US) dataset and 2) the publicly available dataset of cephalometric X-Ray landmark detection, demonstrate the effectiveness of our proposed method. Our proposed framework is general and shows the potential to improve the efficiency of anatomical landmark detection.
BibTeX
@misc{zhou2021learn,
  author = {Guang-Quan Zhou and Juzheng Miao and Xin Yang and Rui Li and En-Ze Huo and Wenlong Shi and Yuhao Huang and Jikuan Qian and Chaoyu Chen and Dong Ni},
  title = {{Learn Fine-grained Adaptive Loss for Multiple Anatomical Landmark Detection in Medical Images}},
  year = {2021},
  eprint = {2105.09124},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2105.09124},
}

Loading related content...

Figure