Topic Title: Predicting the risk of aging associated disease multimorbidity with multimodality data
Population aging is becoming an increasingly important issue around the world. As people live longer, they also tend to suffer from more challenging medical conditions. An elderly population commonly suffers from more than one chronic condition, a situation referred to as multimorbidity. The prevalence and complexity of multimorbidity has been shown to increase in older age groups. It not only decreases quality of life and functional ability, but also puts heavy pressure on healthcare resources. The key to achieve reductions in costs of care is to help societies identify at-risk individuals early in order to provide more effective and timely interventions. AI has been extremely successful in many tasks such as computer vision, speech, or machine translation thanks to improvements in optimization technology and larger datasets. We expect AI-powered systems could also provide affordable tools to diagnose or even predict aging associated diseases. However, due to the complexity of the multimorbidity, the data between different chronic conditions are normally imbalanced, which make classic supervised learning difficult to generalize on those scarce cases. Transfer learning which transfers gained knowledge from source domains and tasks to target domains or tasks enables the trained model to perform well on those diseases with limited labelled data.
- An AI engine utilize multimodality data such as CT, MRIs, or PET images to output predictions of certain aging associated diseases for medical professionals as an assistance .
- The system can be trained in complicated real-world scenarios which requires the trained model has the capability to transfer from the target domain to new domains with limited number of new training data.
- The system has friendly user interfaces, for example some game-based interactions, so that parts of system can be extended from hospitals to communities and homes.
Related Research Topic
Transfer Learning: The traditional supervised learning performs poorly when we do not have sufficient labelled data for the task or domain we care about. If we want to train a model to detect a new disease, we could fine-tune a model that has been trained on a similar disease. In practice, however, we often experience a deterioration or collapse in performance as the model has inherited the bias of its training data and does not know how to generalize to the new domain or task. If we want to train a model to perform a new task, we cannot even reuse an existing model, as the labels between the tasks differ. Transfer learning allows us to deal with these scenarios by leveraging the already existing labelled data of some related task or domain.