Related Experiment Video
Updated: Sep 27, 2025
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
Deep Gaussian processes for multiple instance learning: Application to CT intracranial hemorrhage detection
Miguel López-Pérez1, Arne Schmidt1, Yunan Wu2
1Department of Computer Science and Artificial Intelligence, University of Granada, Granada 18010, Spain.
Insights
This study introduces a Deep Gaussian Process Multiple Instance Learning (DGPMIL) model for detecting intracranial hemorrhage (ICH) in head CT scans. The DGPMIL model accurately diagnoses ICH using only scan-level annotations, reducing the need for costly radiologist input.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Machine Learning
Background:
- Intracranial hemorrhage (ICH) is a critical medical emergency with high mortality and morbidity rates.
- Early and accurate ICH detection is vital for effective patient treatment.
- Current deep learning models for ICH detection often require time-consuming slice-level radiologist annotations.
Purpose of the Study:
- To develop a deep learning model for accurate intracranial hemorrhage detection using only scan-level annotations.
- To address the limitations of costly and time-consuming slice-level annotations in current diagnostic models.
- To improve the efficiency and accuracy of ICH diagnosis in head CT scans.
Main Methods:
- Formulated intracranial hemorrhage detection as a Multiple Instance Learning (MIL) problem.
- Developed a novel probabilistic method using Deep Gaussian Processes (DGP) for MIL training.
- Employed a Convolutional Neural Network (CNN) with an attention mechanism to extract image features, feeding them into the DGPMIL model.
Main Results:
- The proposed DGPMIL model demonstrated superior performance compared to existing MIL methods and attention-based CNNs.
- Experiments on MNIST showed that multiple Gaussian Process layers enhance performance in complex feature distributions.
- DGPMIL achieved high performance on public datasets (RSNA and CQ500), with AUC-ROC of 0.957 and 0.909, respectively.
Conclusions:
- The DGPMIL model provides accurate slice- and scan-level ICH diagnosis without requiring slice-level annotations.
- This approach significantly reduces the annotation burden on radiologists.
- The DGPMIL model's applicability extends to broader medical image classification tasks.
Background And Objective:
Intracranial hemorrhage (ICH) is a life-threatening emergency that can lead to brain damage or death, with high rates of mortality and morbidity. The fast and accurate detection of ICH is important for the patient to get an early and efficient treatment. To improve this diagnostic process, the application of Deep Learning (DL) models on head CT scans is an active area of research. Although promising results have been obtained, many of the proposed models require slice-level annotations by radiologists, which are costly and time-consuming.
Methods:
We formulate the ICH detection as a problem of Multiple Instance Learning (MIL) that allows training with only scan-level annotations. We develop a new probabilistic method based on Deep Gaussian Processes (DGP) that is able to train with this MIL setting and accurately predict ICH at both slice- and scan-level. The proposed DGPMIL model is able to capture complex feature relations by using multiple Gaussian Process (GP) layers, as we show experimentally.
Results:
To highlight the advantages of DGPMIL in a general MIL setting, we first conduct several controlled experiments on the MNIST dataset. We show that multiple GP layers outperform one-layer GP models, especially for complex feature distributions. For ICH detection experiments, we use two public brain CT datasets (RSNA and CQ500). We first train a Convolutional Neural Network (CNN) with an attention mechanism to extract the image features, which are fed into our DGPMIL model to perform the final predictions. The results show that DGPMIL model outperforms VGPMIL as well as the attention-based CNN for MIL and other state-of-the-art methods for this problem. The best performing DGPMIL model reaches an AUC-ROC of 0.957 (resp. 0.909) and an AUC-PR of 0.961 (resp. 0.889) on the RSNA (resp. CQ500) dataset.
Conclusion:
The competitive performance at slice- and scan-level shows that DGPMIL model provides an accurate diagnosis on slices without the need for slice-level annotations by radiologists during training. As MIL is a common problem setting, our model can be applied to a broader range of other tasks, especially in medical image classification, where it can help the diagnostic process.
