Dementia is characterized by the loss of mental functions. Reminiscence therapy, where caregivers interact with patients using their past information (e.g. photos), is the most popular psychosocial intervention in dementia care. Although there are some technological solutions, these are few, and produce a collection of data that remains unchanged, regardless of the emotional reactions throughout sessions. Our goal is to develop a solution that delivers personalized reminiscence therapy at patient's home, being able to adapt the session if the patient experiences negative emotional reactions (using physiological signals). In these situations, our system will select another image to be shown using both emotional information of the images and life information of the patient, with the aim of mitigating negative experiences, while trying to avoid aggressive behaviors that informal caregivers (family) may have to deal with. Our contributions include identifying, developing, and evaluating techniques to identify emotions both from visual stimuli and physiological data.