Authors:
Rémi Besson
1
;
Erwan Le Pennec
1
;
2
;
Emmanuel Spaggiari
3
;
Antoine Neuraz
3
;
Julien Stirnemann
3
and
Stéphanie Allassonnière
4
Affiliations:
1
CMAP, École Polytechnique, Route de Saclay, 91128 Palaiseau, France
;
2
XPop, Inria Saclay, 91120 Palaiseau, France
;
3
Necker-Enfants Malades Hospital, Paris-Descartes University, 149 Rue De Sèvres, 75015 Paris, France
;
4
School of Medicine, Paris-Descartes University, 15 Rue de l’ École de Médecine, 75006 Paris, France
Keyword(s):
Symptom Checker, Stochastic Shortest Path, Decision Tree Optimization, Planning in High-dimension.
Abstract:
In this work, we propose a new optimization formulation for a sequential decision making problem for a rare disease diagnostic application. We aim to minimize the number of medical tests necessary to achieve a state where the uncertainty regarding the patient’s disease is less than a predetermined threshold. In doing so, we take into account the need in many medical applications, to avoid as much as possible, any misdiagnosis. To solve this optimization task, we investigate several reinforcement learning algorithms and make them operable in our high-dimensional setting: the strategies learned are much more efficient than classical greedy strategies.