Index

Gradient-Based Automated Computed Tomography Geometry Correction

Analysis of Federated Learning Approaches for Training Thoracic Abnormality Classification Systems

Project description

Federated learning (FL) [1] is a technique to collaboratively train deep learning (DL) models without the need to host data on a central server. Instead, the training process occurs locally at each participating institution, and only model characteristics, e. g., weights or gradients, are shared to build a consensus model [2]. Thus, as the data remains where it was acquired, FL resolves data governance and ownership issues, making it practical especially in medical imaging where data sharing may be critical due to the presence of personally identifiable information [3]. Nevertheless, FL does not itself guarantee security and privacy since training data may be recovered by reconstruction attacks. Therefore, FL only offers an infrastructural approach to privacy and security, unless combined with other privacy-enhancing techniques (PETs) [4]. Ziegler et al.[5] investigated the feasibility of FL combined with a privacy mechanism called differential privacy (DP) [6, 7] for chest X-ray classification. However, the authors only conduct experiments with one aggregation method and a fixed number of clients.

In this work, we aim to explore FL approaches for the training of reliable thoracic abnormality classification systems. We will investigate different aggregation methods, e. g., [1, 8, 9], to be able to train the classifiers in a federated manner. In doing so, we will focus in particular on analyzing the effect of the number of clients. Furthermore, we will analyze the effect of additionally applied PETs such as DP on model training. We will perform an in-depth evaluation of the performance of the trained classifiers. For our experiments, we will employ the NIH ChestX-ray14 [10] dataset, a collection of 112,120 frontal-view chest X-ray images from 30,805 unique patients with the text-mined fourteen abnormality labels.

The Master’s thesis covers the following aspects:

  1. Overview of the current state-of-the-art for PETs in medical imaging. This overview focuses in particular on FL and DP.
  2. Building a federated pipeline for training thoracic abnormality classification systems which includes:
    • Hyper-parameter tuning
    • Analyzing the effect of different aggregation methods
    • Analyzing the effect of the number of clients
    • Analyzing the effect of DP on model training
  3. Evaluating and comparing the performance of the trained classifiers.
  4. Investigating the feasibility and success of reconstruction attacks (optional).

All DL implementations will be implemented using PyTorch [11].

 

References

[1] Brendan McMahan et al. “Communication-Efficient Learning of Deep Networks from Decentralized Data”. In: Artificial Intelligence and Statistics. PMLR. 2017, pp. 1273–1282.

[2] Nicola Rieke et al. “The future of digital health with federated learning”. In: NPJ Digital Medicine 3.1 (2020), pp. 1–7.

[3] Kai Packhäuser et al. “Deep learning-based patient re-identification is able to exploit the biometric nature of medical chest X-ray data”. In: Scientific Reports 12.1 (2022), pp. 1–13.

[4] Georgios A Kaissis et al. “Secure, privacy-preserving and federated machine learning in medical imaging”. In: Nature Machine Intelligence 2.6 (2020), pp. 305–311.

[5] Joceline Ziegler et al. “Defending against Reconstruction Attacks through Differentially Private Federated Learning for Classification of Heterogeneous Chest X-Ray Data”. In: arXiv preprint arXiv:2205.03168 (2022).

[6] Cynthia Dwork et al. “Calibrating Noise to Sensitivity in Private Data Analysis”. In: Theory of Cryptography Conference. Springer. 2006, pp. 265–284.

[7] Alexander Ziller et al. “Medical imaging deep learning with differential privacy”. In: Scientific Reports 11.1 (2021), pp. 1–8.

[8] Hongyi Wang et al. “Federated Learning with Matched Averaging”. In: arXiv preprint arXiv:2002.06440 (2020).

[9] Tian Li et al. “Federated Optimization in Heterogeneous Networks”. In: Proceedings of Machine Learning and Systems 2 (2020), pp. 429–450.

[10] XiaosongWang et al. “ChestX-ray8: Hospital-Scale Chest X-Ray Database and Benchmarks on Weakly-Supervised Classification and Localization of Common Thorax Diseases”. In: The IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2017.

[11] Adam Paszke et al. “PyTorch: An Imperative Style, High-Performance Deep Learning Library”. In: Advances in Neural Information Processing Systems 32 (2019), pp. 8026–8037.

Joining photogrammetry and X-Ray computed Tomography

Edge-AI: Self-sensing backpressure estimation in piezoelectric micropumps using machine learning methods on a limited hardware

Advancing Classification and Segmentation of Gastrointestinal Diseases by Leveraging Large Unlabeled Datasets

Uncertainty Estimation for Transformer-based Glacier Segmentation

Deep Learning-Based Optical Flow for Camera Pose Estimation in Navigation Assistance for Blind Pedestrians on Sidewalks

Automated knowledge management using a semantic database and large language models in the medical domain

Deep Metric Learning for Orca Identification

A hybrid approach forLeakage Localization in the Water Distribution Network

Climate change is expected to cause more frequent and intense weather events such as droughts and floods, which can place additional stress on water distribution networks (WDN). Leakage in water distribution networks is a significant challenge that exacerbate the effects of climate change by increasing the amount of water that needs to be extracted and treated, as well as increasing energy consumption and greenhouse gas emissions associated with pumping and treating water. Therefore, accurate leakage localization can help reduce the amount of water lost from distribution networks, thereby reducing the need for additional water extraction and treatment. This can lead to energy savings and reduced greenhouse gas emissions, as well as ensuring that water resources are used efficiently and effectively. Additionally, by reducing the amount of water lost to leakage, WDN can be made more resilient to the impacts of climate change, such as droughts and water scarcity.
State-of-the-art methods tackle the challenge of leakage localization in a WDN comprise acoustic methods [1], pressure transient methods [2], flow measurement methods [3], and machine learning (ML) based methods [4-5]. However, these methods have significant limitations that hinder their application in the daily routine of water utilities. For instance, acoustic methods are cost intensive as they require additional sensors and equipment. Furthermore, the accuracy of such methods is greatly affected by the material of the pipes and presence of noise. Although the sensors that are necessary for the pressure transient methods and flow measurement methods might available due to the daily operation of WDN, these methods are often not sensitive to detect small leaks. Data-driven methods using ML has gain more importance in the recent year. However, the data availability, data quality and the explainability of ML models are the major limitations.
Therefore, we would like to investigate the effectiveness of a hybrid AI approach combining hydraulic model and ML to tackle the leakage localization within WDN using real world data. The following aspects need to be considered:
• Literature review of leakage localization for WDN.
• Development and implementation of a hybrid framework combining hydraulic model and ML methods for leakage localization with existing sensor data.
• Comprehensive evaluation of the performance of the implemented framework w.r.t. accuracy, robustness, and explainability.
[1] Khulief, Yehia et. al., (2012). Acoustic Detection of Leaks in Water Pipelines Using Measurements inside Pipe. ASCE Journal of Pipeline System Engineering and Practice. 2021, 3, 47. Doi:10.1061/(ASCE)PS.1949-1204.0000089.
[2] Levinas, D. et. al., Water Leak Localization Using High-Resolution Pressure Sensors. Water 2021, 13, 591. https://doi.org/10.3390/w13050591
[3] L. Lindström, et. al. Leakage Localization in Water Distribution Networks: A Model-Based Approach, 2022 European Control Conference (ECC), London, United Kingdom, 2022, pp. 1515-1520, doi: 10.23919/ECC55457.2022.9838006.
[4] Huang, Pingjie, et al. “Real-time burst detection in district metering areas in water distribution system based on patterns of water demand with supervised learning.” Water 10.12 (2018): 1765. doi.org/10.3390/w10121765
[5] Soldevila, Adrià, et al. “Data-driven approach for leak localization in water distribution networks using pressure sensors and spatial interpolation.” Water 11.7 (2019): 1500. doi.org/10.3390/w11071500