Improved Planning Model for Patient

Note: This application has been filled. Background Individualizing ovarian stimulation and optimizing workflow in in vitro fertilization (IVF) remains a major clinical and logistical challenge. Treatment outcome and stimulation length are influenced by a wide range of patient-specific factors, including age, ovarian reserve, body mass index (BMI), hormonal profiles, stimulation protocol, medication dose, and previous cycle response. Despite the use of different stimulation protocols based primarily on age and ovarian reserve, the actual duration of stimulation remains unpredictable in approximately 40% of IVF cycles. This unpredictability has significant consequences for both patients and clinical workflows. Patients often require multiple ultrasound visits to monitor follicular development and adjust medication, increasing treatment burden and stress. At the same time, IVF laboratory workflows are affected by large fluctuations in the number of oocyte pick-ups (OPUs) per day, ranging from one to ten procedures. Such variability complicates staffing, resource allocation, and scheduling, and may ultimately affect treatment efficiency and outcomes. Currently, no data-driven approach is available to accurately predict stimulation length and the timing of ovum pick-up at the start of treatment. The absence of predictive models limits the ability to provide patient-centered planning and results in inefficient use of clinical and laboratory resources. Recent work has demonstrated that IVF laboratory performance and workflow characteristics are associated with treatment outcomes, highlighting the potential value of predictive planning tools to improve both efficiency and quality of care [1]. Approach The objective of this project is to develop an AI-driven predictive model that estimates the timing of ovum pick-up for individual IVF patients with an accuracy of one to two days. The model will integrate heterogeneous data sources, including baseline patient characteristics (e.g. age, BMI, ovarian reserve markers), hormonal profiles, stimulation protocol, medication type and dose, and outcomes from previous cycles. Using historical IVF cycle data, the model will learn patterns that relate early-cycle characteristics to stimulation duration and optimal OPU timing. The prediction will be generated at the start of ovarian stimulation and updated as needed, enabling improved planning of patient monitoring, laboratory workflow, and staffing. The project will focus on developing interpretable and clinically usable outputs that can support decision-making while minimizing risks such as ovarian hyperstimulation syndrome (OHSS) or poor response. Data The dataset consists of approximately 5,000 IVF cycles recorded between 2016 and the present at Radboudumc, covering IVF, ICSI, and ICSI-TESE treatments. The data include both baseline patient characteristics and longitudinal treatment variables, such as medication use, hormonal measurements, ultrasound findings, number and quality of retrieved oocytes, embryo quality, and pregnancy outcomes. References [1] Innocenti F, Cermisoni GC, Taggi M, et al. Optimizing IVF lab workflows through data-driven insights: associations between lab management, procedural timings, and workload with blastulation rates. Human Reproduction. 2025;40(11):. Requirements Students in the final phase of a Master's program in Artificial Intelligence, Data Science, Biomedical Engineering, Computer Science, or a related field are invited to apply. Required skills: • Experience with Python programming • Familiarity with machine learning and data analysis