BA/MA: PET-Histology Prostate Cancer Segmentation
Advisor: Prof. Bjoern Menze
Supervision by: Fernando Navarro
In this project, prostate cancer segmentation will be studied leveraging PET and Histology modalities in a voxel to voxel correspondence. The student will apply state-of-the-art deep learning architectures for medical image segmentation.
Figure 1: Visual Evaluation of spatial correlation between PSMA-PET and histopathology.
Left column: axial CT images, middle column: corresponding axial PSMA-PET images,
right column: corresponding histopathological slices .
Background and Motivation
Detecting prostate cancer is still a challenging task in medical routine even for experienced physicians, mainly because of the variability of shape, texture, and deformability of the organs.
The current work-flow for accurate cancer detection is to scan the patient using a PET-CT scanner.
Nevertheless, this procedure can lead to inaccurate results, due to the lack of ground truth for example biopsy or histology.
This means that PET-CT cancer detection is still a subjective task that depends on the physician expertise and the assumed registration between PET and CT .
Aiming to solve the aforementioned problems, we have collected a dataset of PET-CT-Histology modalities in a voxel to voxel correspondence .
This is the first work in this direction, where histology is used as the absolute ground truth for every voxel in the other modalities.
The aim of this project is to develop and implement deep neural networks for image segmentation in an end-to-end strategy for PET-Histology.
The questions to answer is whether we can make predictions-like histology based on PET or CT alone or what is the correlation between Histology and PET.
The student task would be to perform the following tasks:
- Develop and implement deep neural networks to automatically predict
prostate cancer segmentation in PET-CT based on an in-house annotated dataset (100 scans)
- Leverage the collected PET-Histology dataset (20 scans) to develop and implement state-of-the-art
deep learning architectures to predict histology-like segmentation.
The student should have prior knowledge in :
- Image processing
- Experience with deep learning
- Good python programming skills
- Experience with deep learning frameworks like TensorFlow? or Pytorch
 Calais, J., Czernin, J., Cao, M., Kishan, A.U., Hegde, J.V., Shaverdian, N., Sandler, K., Chu, F.I., King, C.R., Steinberg, M.L. and Rauscher, I., 2018. 68Ga-PSMA-11 PET/CT mapping of prostate Cancer biochemical recurrence after radical prostatectomy in 270 patients with a PSA level of less than 1.0 ng/mL: impact on salvage radiotherapy planning. J Nucl Med, 59(2), pp.230-237.
 Zamboglou, C., Schiller, F., Fechter, T., Wieser, G., Jilg, C.A., Chirindel, A., Salman, N., Drendel, V., Werner, M., Mix, M. and Meyer, P.T., 2016. 68Ga-HBED-CC-PSMA PET/CT versus histopathology in primary localized prostate cancer: a voxel-wise comparison. Theranostics, 6(10), p.1619.