Index

Deep Learning–Driven Lorentzian Fitting for 31P Spectrum

To isolate different peaks in the phosphorus spectrum, several preprocessing steps are usually performed, and the final information about the different metabolites is extracted by multiple Lorentzian line fitting of the spectrum [1]; this least-squares fitting is prone to noise and also depends on preprocessing steps [2]. This study will investigate the use of the Lorentzian distribution generator as a known operator in a fully connected network to fit the phosphorus spectrum.

The thesis will include the following points:

1. Number of Lorentzian distributions required to fit the phosphorus spectrum
2. Comparison of the least square fit and the deep Lorentzain fit
3. Correlation of deep lorentzain fit peaks with tumor types

References:

1. Meyerspeer M, Boesch C, Cameron D, et al. 31 P magnetic resonance spectroscopy in skeletal muscle: Experts’ consensus recommendations. NMR Biomed. Published online February 10, 2020. doi:10.1002/nbm.4246
2. Rajput, J.R., et al.: Physics-informed conditional autoencoder approach for robust metabolic CEST MRI at 7T. In: Greenspan, H., et al. Medical Image Computing and Computer Assisted Intervention – MICCAI 2023. MICCAI 2023. LNCS, vol. 14227. Springer, Cham (2023)

Pathology detection in medical images

This work will investigate applying computer vision detection techniques in medical images

Required: strong skills in

• proogramming python
• deep learning , training methods, pattern recognition, loss functions
• medical imaging background – X-rays, CT scan images, DICOM processing
• communication, scientific writing

Low Field MR Image Denoising

Design of an improved automatic exposure control algorithm in mammography using breast shape and tissue information

Influence of Demographic Parameters in Radar-Based Blood Pressure Estimation

Continuous non-invasive blood pressure (BP) monitoring is a critical advancement in healthcare,
allowing for the early detection and management of hypertension one of the leading risk factors
for cardiovascular diseases [1, 2, 3]. Traditional cuff-based methods, though widely used, provide
only intermittent readings and fail to capture BP fluctuations throughout the day. To address
this limitation, machine learning models leveraging radar-based skin displacement signals have
been proposed for continuous BP estimation. However, the influence of demographic factors
such as age, gender, height, and weight on prediction accuracy remains insufficiently explored.
Existing studies primarily focus on machine learning models trained without explicit demographic
considerations. While some methods integrate features such as Pulse Transit Time
(PTT) from Photoplethysmography (PPG) sensors [4, 5, 6, 7], fewer approaches investigate the
role of demographic characteristics in radar-based BP monitoring [8]. This research aims to systematically
analyze the impact of demographic features on BP prediction using a Transformer-based
deep learning model.
The current methodology relies on pretraining a model on a large-scale arterial blood pressure
dataset (PulseDB [9]) and fine-tuning it with radar-based BP measurements [3] from human
participants. An ablation study is conducted to assess the contribution of individual demographic
features to model performance. Evaluation metrics include the Mean Absolute Error
(MAE) and Standard Deviation (STD), as well as compliance with established medical standards
such as those from the Association for the Advancement of Medical Instrumentation
(AAMI) [10] and the British Hypertension Society (BHS) [11].

The main objectives of this thesis include:

• Analyze the impact of demographic characteristics on BP prediction accuracy.
• Compare the performance of two distinct model architectures: a feed-forward neural
  network and a transformer network.

To achieve these objectives, the following proposed steps will be undertaken:

1. Analyze and Compare the distribution of the demographic features in both datasets
2. retrain both model architectures on a large-scale arterial blood pressure dataset
3. Finetune both model architectures on the radar dataset
4. Evaluate and assess the model’s performance against set criteria.

References
[1] X. Xing, Z. Ma, M. Zhang, Y. Zhou, W. Dong, and M. Song, “An Unobtrusive and
Calibration-free Blood Pressure Estimation Method using Photoplethysmography and Biometrics,”
vol. 9, no. 1, p. 8611.
[2] D. Barvik, M. Cerny, M. Penhaker, and N. Noury, “Noninvasive Continuous Blood Pressure
Estimation From Pulse Transit Time: A Review of the Calibration Models,” vol. 15,
pp. 138–151.
[3] N. Vysotskaya, C. Will, L. Servadei, N. Maul, C. Mandl, M. Nau, J. Harnisch, and A. Maier,
“Continuous Non-Invasive Blood Pressure Measurement Using 60 GHz-Radar—A Feasibility
Study,” vol. 23, no. 8, p. 4111.
[4] S. González, W.-T. Hsieh, and T. P.-C. Chen, “A benchmark for machine-learning based
non-invasive blood pressure estimation using photoplethysmogram,” vol. 10, no. 1, p. 149.
[5] R. Mukkamala, J.-O. Hahn, O. T. Inan, L. K. Mestha, C.-S. Kim, H. Töreyin, and S. Kyal,
“Toward Ubiquitous Blood Pressure Monitoring via Pulse Transit Time: Theory and Practice,”
vol. 62, no. 8, pp. 1879–1901.
[6] S. Maqsood, S. Xu, M. Springer, and R. Mohawesh, “A Benchmark Study of Machine
Learning for Analysis of Signal Feature Extraction Techniques for Blood Pressure Estimation
Using Photoplethysmography (PPG),” vol. 9, pp. 138817–138833.
[7] C. El-Hajj and P. Kyriacou, “A review of machine learning techniques in photoplethysmography
for the non-invasive cuff-less measurement of blood pressure,” vol. 58, p. 101870.
[8] N. Vysotskaya, N. Maul, A. Fusco, S. Hazra, J. Harnisch, T. Arias-Vergara, and A. Maier,
“Transforming Cardiovascular Health: A Transformer-Based Approach to Continuous,
Non-Invasive Blood Pressure Estimation via Radar Sensing,” in ICASSP 2024 – 2024
IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP),
pp. 2041–2045.
[9] W. Wang, P. Mohseni, K. L. Kilgore, and L. Najafizadeh, “PulseDB: A large, cleaned
dataset based on MIMIC-III and VitalDB for benchmarking cuff-less blood pressure estimation
methods,” vol. 4, p. 1090854.
[10] G. S. Stergiou, B. Alpert, S. Mieke, R. Asmar, N. Atkins, S. Eckert, G. Frick, B. Friedman,
T. Graßl, T. Ichikawa, J. P. Ioannidis, P. Lacy, R. McManus, A. Murray, M. Myers,
P. Palatini, G. Parati, D. Quinn, J. Sarkis, A. Shennan, T. Usuda, J. Wang, C. O. Wu, and
E. O’Brien, “A Universal Standard for the Validation of Blood Pressure Measuring Devices:
Association for the Advancement of Medical Instrumentation/European Society of Hypertension/
International Organization for Standardization (AAMI/ESH/ISO) Collaboration
Statement,” vol. 71, no. 3, pp. 368–374.
[11] E. O’Brien, “Blood pressure measuring devices: Recommendations of the European Society
of Hypertension,” vol. 322, no. 7285, pp. 531–536.

Enhancing Breast Abnormality Detection on Mammograms with Advanced Vision-Language Models

Pre-processing and synthetic data generation techniques for defect detection in SiC and AlN wafers

Deep Learning-Based Fault Detection and Classification in Power System Protection: A Comparative Study

Power systems are susceptible to faults during electricity transmission, which can compromise system stability and reliability. Accurate fault detection and classification are crucial for implementing effective protection and mitigation strategies. While deep learning models have demonstrated considerable potential for automating these tasks, their performance can vary depending on the specific circumstances. This research provides a comparative analysis of various deep learning approaches for fault detection and classification in power system protection. The models are assessed using a relevant dataset and multiple performance metrics to gauge their effectiveness. The aim is to offer a structured evaluation, providing insights into which approaches are best suited for different fault scenarios.

Establishment of a Fully Automated Treatment Planning Pipeline for Prostate Cancer Brachytherapy (Radiotherapy)

Radiation therapy, specifically brachytherapy, is one of the most important local cancer treatment modalities for prostate cancer, enabling the delivery of a spatially varying dose distribution within a patient’s body with very high precision [1]. Brachytherapy treatment planning aims to identify the optimal treatment plan that maximally spares surrounding normal tissue structures while delivering a high dose to the target volume. For this process, a sufficient delineation of target volume, organs at risk, and identification of implanted brachytherapy needles in ultra-sound images is crucial. Deep Learning segmentation for brachytherapy treatment planning in ultrasound images is still a challenging problem, that could automate and smoothen the treatment planning process, improve standardization and reduce the treatment planning time to enable novel and improved therapies like adaptive brachytherapy [2]. Auto-segmentation of brachytherapy ultra-sound images is challenging, as ultra-sound images are noisy, have high slice-thickness and contain artifacts (e.g. from implanted needles). Based on deep learning auto-segmentations, dose calculation and optimization algorithms can be used to automatically create deliverable treatment plans [3]. The aim of this master thesis project is to create a deep learning auto-segmentation pipeline for brachytherapy ultrasound images to automate treatment planning on a cohort of more than 600 prostate cancer datasets from the Department of Radiation Oncology, University Hospital Erlangen.

Figure 1: Overview of target volume and organ segmentation as well as needle reconstruction as performed during clinical routine (left). Structure delineation on acquired ultrasound images enables planning of the treatment dose distribution (right). Image from Karius et al. [4].

The thesis will include the following points:

• Literature review on deep learning techniques for target volume and organs-at-risk segmentation, as well needle detection on prostate Literature review on novel techniques to improve auto-segmentation perfor- mance on noisy ultra-sound images with artifacts including cardiac and fetal ultra-sound that could be trans- ferred to brachytherapy ultrasound auto-segmentation with a particular focus on integrating prior knowledge on structure shape.
• Data preprocessing for the private ultrasound treatment plan dataset including conversion of DICOM (CT, RT STRUCT) files into label maps. Automatic cleaning and splitting of the datasets into subgroups according to clinical treatment concepts using custom Python scripts (Pydicom library).
• Conversion of available needle point coordinates into binary segmentations by mapping of the known needle shape to the point coordinates.

• Training, Inference, and Testing of an nnU-net 2D and 3D fullres model on the cohort of 600 prostate can- cer ultrasound datasets to establish a baseline for comparison. Comparison of the ultrasound segmentation performance for the 2D and 3D model variant.
• Improve segmentation performance by integrating shape priors into ultrasound segmentation networks. Inte- gration of the novel method of shape prior modules (SPMs [5]) into a Unet-based segmentation Com- parative evaluation of SPM with the conventional U-net model.
• Fallback alternative / optional: Evaluate a Vision Transformer (e.g. Nnformer [6]) or a CNN-Transformer- Hybrid for brachytherapy ultrasound auto-segmentation, which could provide improved ultrasound segmenta- tion performance by improved modeling of global image context.
• Establish a pipeline for processing clinical brachytherapy ultra-sound images in DICOM format, auto-segmenting the images and exporting the auto-segmentations into DICOM RTStruct format for import into routine treatment planning programs. Evaluation of the pipeline processing time.
• Detailed evaluation of the developed brachytherapy ultra-sound auto-segmentation solution in reference to clin- ical ground truths. Perform five training and inference repeats to enable statistical analysis.

If you are interested in the project, please send your request to: andre.karius@uk-erlangen.de

Having prior experience with building neural networks in Python, especially using frameworks such as PyTorch or TensorFlow, will greatly help to develop the project.

Category-Level Segmentation of industrial Parts Using SAM2 Memory System

The Segment Anything Model (SAM), developed by Meta AI in 2023, introduced a powerful zero-shot segmentation approach, allowing object segmentation without additional training. In 2024, Meta released SAM2, an upgraded version designed for video object tracking using a hierarchical vision transformer and a cross-attention mechanism with memory integration. Despite its advancements, SAM2 lacks category-specific segmentation, limiting its ability to distinguish objects based on contextual understanding.

This research aims to enhance SAM2 by leveraging its automatic mask generator (AMG) and memory system to improve object recognition. The proposed method involves generating segmentation candidates using AMG and employing cross-attention mechanisms to compare feature similarities, enabling consistent identification of objects across frames. Additionally, to address SAM2’s challenges with light reflection and over-segmentation, small-scale fine-tuning of the image encoder and mask decoder will be implemented.

The study will include an analysis of SAM2’s architecture, the development of an embedding-based object identification method, and performance benchmarking using the ARMBench dataset and a custom industrial dataset. The findings will contribute to improving SAM2’s capability in complex environments, particularly in robotic applications requiring precise object classification and tracking.