PERSONALISED MEDICINE: MULTIDISCIPLINARY RESEARCH TOWARDS IM (PersoRad)

Prostate cancer (PCa) is the most frequent diagnosed malignancy in male patients and radiation therapy (RT) is a main treatment option. However, RT concepts for PCa have an imminent need to be rectified in order to personalise the RT strategy by considering (i) the individual PCa biology and (ii) the individual disease process of each patient. The consortium concatenates a prospective, non-randomized phase II trial for personalised RT of PCa patients (HypoFocal) with novel tools for patient involvement, advanced omic and bioinformatical analyses. In detail: (i) data comprising clinical course and individual PCa radio resistance (measured by yH2AX foci in tissue specimen) will be interpolated by state of the art imaging techniques (PSMA PET/CT and mpMRI) as a radiomics approach and with genomic profiling of the respective tumor tissue samples to identify biomarkers predictive of radio resistance. Additionally, (ii) mobile health tools will be offered to eligible patients of the HypoFocal trial in addition to established health services research approaches, such as interviews and focus groups, in order to increase patient involvement and to obtain patient reported outcomes. Artificial neural network approaches will be employed to support the results which will be implemented in (radio)biological model systems in order to calculate the tumor control probability and normal tissue complication probability of each patient in the future. New online tools with controlled access will address data exchange between project partners ensuring the highest level of data protection. Hereby, the implementation of novel computational components enables (i) unbiased characterization of tumor biology for personalised treatment planning processes (ii) direct integration of patients preferences for a personalised follow-up process after RT and (iii) characterization of environmental and personal context factors that are the basis for these preferences.

Prostate cancer is one of the most common diseases diagnosed in men. One of the main treatment options is using radiotherapy (RT). Efficient RT requires accurate location of the tumour and organs at risk (OAR). However, manual segmentation of organs and intraprostatic gross tumour volumes (GTV) based on MRI and PSMA PET/CT underlies great interobserver heterogeneity and is time-consuming. Thus, artificial intelligent algorithms were implemented to evaluate the performance of automatic GTV and OAR segmentations using convolutional neural networks (CNN). First, a literature review and investigation of different CNN-based U-Net variants was conducted to determine the best performing network on prostate segmentation tasks. Five U-Net variants were evaluated and benchmarked using a standardised framework on four prostate datasets with CT and MR modalities. The results indicated that significant differences in network performance occur when dealing with small-sized datasets. This work was published in Computerized Medical Imaging and Graphics journal in July 2023. Further, a new variant of the well known 2D and 3D U-Nets architectures, called IB-U-Nets was developed, taking advantage of an inductive bias inspired by the vertebrate retina, in order to considerably improve their discrimination abilities. In a first approach, the IB-U-Nets were trained on prostate data alone, and the results were published in the AAAI 2022 Workshop on Trustworthy AI for Healthcare. In the later stages, to improve the segmentation precision, IB-U-Nets were trained to segment multiple organs, including the prostate and the bladder from CT and MR volumes. The resulting IB-U-Nets were shown to be robust against multiple artefacts and out-of-distribution samples. Finally, IB-U-Nets were also trained to locate tumour volumes from PSMA-PET images. The best performing models were used to create a 3D Slicer extension that could be used by healthcare workers without additional inputs. The experiments and results of IB-U-Nets are under review in the Medical Image Analysis journal.