Tinnitus is an annoying ringing in the ears, in varying shades and intensities. Tinnitus can affect a person's overall health and social well-being (e.g., sleep problems, trouble concentrating, anxiety, depression and inability to work). The diagnostic procedure of tinnitus usually consists of three steps: an audiological examination, psychoacoustic measurement, and a disability evaluation. All steps are performed by physicians, who use specialised hardware/software and administer questionnaires. This paper presents a system, to be used by patients, for the diagnosis and self-management of tinnitus. The system is made up of an app and a device. The app is responsible for executing - through the device - a part of the required audiological and psychoacoustic examinations, as well as administering questionnaires that evaluate disability. The paper reviews the quality of the automated audiometric reporting and the user experience provided by the app. Descriptive and inferential statistics were used to support the findings. The results show that automated reporting is comparable with that of physicians and that user experience was improved by re-designing and re-developing the acufenometry of the app. As for the user experience, two experts in Human-Computer Interaction evaluated the first version of the app: their agreement was good (Cohen's K = 0.639) and the average rating of the app was 1.43/2. Also patients evaluated the app in its initial version: the satisfactory tasks (audiometry and questionnaires) were rated as 4.31/5 and 4.65/5. The unsatisfactory task (acufenometry) was improved and the average rating increased from 2.86/5 to 3.96/5 (p = 0.0005). Finally, the general usability of the app was increased from the initial value of 73.6/100 to 85.4/100 (p = 0.0003). The strengths of the project are twofold. Firstly, the automated reporting feature, which - to the best of our knowledge - is the first attempt in this area. Secondly, the overall app usability, which was evaluated and improved during its development. In summary, the conclusion drawn from the conducted project is that the system works as expected, and despite some weaknesses, also the replication of the device would not be expensive, and it can be used in different scenarios.

Deep learning models achieve high-quality results in image processing. However, to robustly optimize parameters of deep neural networks, large annotated datasets are needed. Image annotation is often performed manually by experts without a comprehensive tool for assistance which is time- consuming, burdensome, and not intuitive. Using the here presented modular Karlsruhe Image Data Annotation (KaIDA) tool, for the first time assisted annotation in various image processing tasks is possible to support users during this process. It aims to simplify annotation, increase user efficiency, enhance annotation quality, and provide additional useful annotation-related functionalities. KaIDA is available open-source at https://git.scc.kit.edu/sc1357/kaida.

Machine learning (ML) is increasingly being used to guide drug discovery processes. When applying ML approaches to chemical datasets, molecular descriptors and fingerprints are typically used to represent compounds as numerical vectors. However, in recent years, end-to-end deep learning (DL) methods that can learn feature representations directly from line notations or molecular graphs have been proposed as alternatives to using precomputed features. This study set out to investigate which compound representation methods are the most suitable for drug sensitivity prediction in cancer cell lines. Twelve different representations were benchmarked on 5 compound screening datasets, using DeepMol, a new chemoinformatics package developed by our research group, to perform these analyses. The results of this study show that the predictive performance of end-to-end DL models is comparable to, and at times surpasses, that of models trained on molecular fingerprints, even when less training data is available. This study also found that combining several compound representation methods into an ensemble can improve performance. Finally, we show that a post hoc feature attribution method can boost the explainability of the DL models.