Radiotherapy is next to chemotherapy and surgery one of the most important treatment methods in cancer therapy. Its aim is to irradiate the tumor with a lethal dose of ionizing irradiation while keeping the side effects on the healthy tissue at a minimum level. There are several approaches improving radiotherapy and reducing the side effects. A very promising approach is the irradiation with ultra-high dose rates (> 40 Gy/s) known as FLASH. FLASH-radiotherapy provides significant sparing of healthy tissue while maintaining tumor control. This effect has been seen in several animal models and applies for protons, electrons, and x-rays. Although this effect has been highly investigated, the underlying mechanisms are still unclear. Therefore, further investigations are necessary. FLASH radiation therapy with protons has three major advantages over other types of irradiation. Due to their characteristic energy distribution in the tissue, protons often lead to better irradiation results than the frequently used x-rays, existing proton beam facilities can be converted to FLASH, and protons, in contrast to the electrons frequently used in the FLASH experiments, enable the irradiation of deep-seated tumors. In this work, I investigated the proton-FLASH effect using three in-vitro models and an in-vivo mouse ear model. Furthermore, a new method based on longterm live-cell phase-contrast microscopy and artificial intelligence based analysis of the obtained videos to investigate several radiobiological effects in-vitro in one assay was developed and presented here. The proton-FLASH experiments were conducted on the SNAKE beamline at the tandem accelerator of the Maier-Leibnitz-Laboratorium in Garching near Munich. As in-vitro experiments, a colony forming assay, a Caspase3/7-Sytox assay and a micronuclei test were performed with doses of 4 Gy and at three different dose rates (Conv = 0.06 Gy/s, Flash9 = 9.3 Gy/s and Flash310 = 310 Gy/s). The colony forming assay investigated the cell survival, the Caspase3/7-Sytox assay the cell death and the micronuclei test the genetic damage. Dose rate effects were observed in a reduction of cell death at a dose rate of 310 Gy/s and in reduced genetic damage at dose rates of 310 Gy/s and 9.3 Gy/s. No difference was seen in the cell survival. In the in-vivo mouse ear model, the right ears of Balb/c mice were irradiated at three dose rates (Conv = 0.06 Gy/s, Flash9 = 9.3 Gy/s and Flash930 = 930 Gy/s) for a total dose of 23 Gy or 33 Gy. Desquamation and erythema of the irradiated ear, which were combined into an inflammation score, and the ear thickness were measured for 180 days. The cytokines TGF-β1, TNF-α, IL1α, and IL1β were analyzed in blood samples collected during the first 4 weeks and on the last day of treatment. No differences in inflammatory responses were observed in the 23 Gy group for the different dose rates. In the 33 Gy group, ear swelling and inflammation score were reduced for Flash9 and Flash930 compared to the Conv dose rate. No changes in the cytokines were measured in the blood. However, an estimation of the irradiated blood volume demonstrated, that 100 times more blood is irradiated when using Conv compared to using Flash9 or Flash930. This suggests that the irradiated blood volume may play a role in the underlying mechanisms of the FLASH effect. In the novel in-vitro method presented in this work, cells were moved directly after irradiation to an inverted phase-contrast microscope equipped with a live-cell set-up that allows recording of living cells for several days. The obtained videos were evaluated by the self-developed artificial intelligence based algorithm CeCILE (Cell Classification and In-vitro Lifecycle Evaluation). CeCILE was trained on a custom dataset I created. To enable the evaluation of the cell cycle, cells in the dataset were assigned to one of four morphological classes, which describe different states in the cell cycle and cell death. The whole development and training process is presented in this thesis. To show CeCILE's potential, my new method was applied to a sample of irradiated cells (3 Gy of x-rays) and a sample of sham irradiated cells. It could be shown that CeCILE sucessfully created cell lineages of every cell in the obtained videos and evaluated the following endpoints: number of cells per frame in the four morphological classes, time points of the first cell divisions of every cell, cell cycle duration, cell cycle abnormalities and proliferation. Therefore, CeCILE is able to accurately assess several important radiobiological endpoints. In this thesis, new insights into the proton-FLASH effect could be gained and more are expected when the novel method based on CeCILE is applied to proton-FLASH in-vitro experiments in the future.