ADS & World AI Summit | Using AI to support patient treatment strategy selection
For this webinar ADS has invited three PhD students to showcase their work and research on Using AI to support patient treatment strategy selection.
16:30 Talk #1 : Evolutionary Intelligent Bi-objective Treatment Planning for Prostate Cancer
16:50 Talk #2: AI-driven optimal medical treatment protocol for mechanically ventilated patients with COVID-19 pneumonia
17:10 Talk #3 : Predicting mortality of individual COVID-19 patients: Can we improve decision making?
Talk #1 by Anton Bouter
Anton Bouter is a PhD candidate in the Life Sciences and Health group at Centrum Wiskunde & Informatica (CWI), Amsterdam, The Netherlands
High-Dose Rate (HDR) brachytherapy is a type of internal cancer treatment. In prostate HDR brachytherapy, the target volumes (prostate, seminal vesicles) should receive a sufficient dose, while organs at risk (bladder, rectum, urethra) should receive as little dose as possible. Finding the best possible treatment plan is a difficult optimization problem, with an inherent trade-off between target coverage and organ sparing. For this reason, BRIGHT was developed. BRIGHT is a novel treatment planning method based on the model-based evolutionary algorithm GOMEA, and is currently used in clinical practice at the Amsterdam UMC, location AMC. BRIGHT produces a large set of high-quality treatment plans with different trade-offs between coverage and sparing, and requires no more than a few minutes due to GPU-acceleration. Resulting plans were evaluated by a number of clinical experts, and were found to be clinically superior to previously used clinical plans.
Talk #2 by Luca Roggeveen
Luca Roggeveen (MD), is a resident at the intensive care unit at the Amsterdam University medical Center, location VUmc.
Hypoxemic acute respiratory failure due to COVID-19 often requires mechanical ventilation, including an approach for setting positive end-expiratory pressure (PEEP) and inspiratory oxygen fraction (FiO2). The easiest solution is the use of PEEP/FiO2 tables. However, lung recruitability is variable and a personalized approach would likely be beneficial. They created a MDP that includes vital signs, laboratory markers and ventilator and settings into a state space. We used a PEEP versus FIO2 table to devise an action space. Least-squares policy iteration (LSPI) and Deep Q Networks (DQN) were used to develop an optimal policy. Off policy evaluation showed the optimal policy has higher then physician performance. A deep policy inspection of the difference between the optimal and physician policy showed potentially clinically actionable differences, indicating room for improvement in current clinical practice.
Talk #3 by Lucas Ramos
Luca is a PhD at the department of Biomedical Engineering and Physics, Amsterdam University Medical Centers (Amsterdam UMC).
The first wave of the COVID-19 pandemic had a dramatic effect on our society and severely disrupted our daily lives, economies and healthcare systems. During the peak of the first wave, hospitals and intensive care units (ICU) throughout Europe were overwhelmed and resources were exhausted. Given the novelty of the virus, accurate information about the clinical course and prognosis of individual patients is still largely unknown, which led to the use of crude limits to withhold advanced life support measures to face the large numbers of pulmonary insufficient patients during the first wave. Although criticized, several hospitals in Europe have already solely used age as a triage criterion. In this study we developed machine learning models using data from a multicenter Dutch cohort of Covid-19 patients, to predict 21-day mortality, and found new insights that can assist decision making during a crisis situation.
You can find the full programme of the day on the Inspired AI website.
Free tickets to this events are available!