Use Artificial Intelligence methods to learn from health data
Clinical Data Science uses artificial intelligence methods to learn from health and health related data.
We apply a wide variety of machine and deep learning algorithms. We use AI for two main goals:
Improve efficiency and/or quality of health care
Efficiency and quality of health care can be improved if AI applications can perform tasks that either take a lot of human resources and/or need specialized skills. Deep learning for instance can perform routine tasks of radiation oncologists and technologists such as 3D segmentation of tumors and normal tissues in images with a performance that matches or surpasses human performance. And Natural Language Processing can make it easier to extract concepts from medical free text and documents.
Support decisions to improve future health
Prediction of a future health outcomes of an individual, like survival in a cancer patient after radiotherapy, is a difficult task as such outcomes are usually determined by many factors. But not being able to predict the future limits our ability to recommend treatments and other interventions (such as a lifestyle change) to individual patients and citizens. AI can handle many more predictive factors than humans and we thus do research in AI driven outcome prediction models. We use machine learning on real world data to derive such models. Determining validity, bias and generalizability of health outcome prediction models is a particular strength of our group. We also focus on relative simple and transparent AI as treatment decisions need to be explained and understood.
Publications in this theme
Using 3D deep features from CT scans for cancer prognosis based on a video classification model: A multi-dataset feasibility study
Background
Cancer prognosis before and after treatment is key for patient management and decision making. Handcrafted imaging biomarkers—radiomics—have shown potential in predicting prognosis.
Purpose: However, given the recent progress in deep learning, it is timely and relevant to pose the question: could deep learning based 3D imaging features be used as imaging biomarkers and outperform radiomics?
Methods: Effectiveness, reproducibility in test/retest, across modalities, and correlation of deep features with clinical features such as tumor volume and TNM staging were tested in this study. Radiomics was introduced as the reference image biomarker. For deep feature extraction, we transformed the CT scans into videos, and we adopted the pre-trained Inflated 3D ConvNet (I3D) video classification network as the architecture. We used four datasets—LUNG 1 (n = 422), LUNG 4 (n = 106), OPC (n = 605), and H&N 1 (n = 89)—with 1270 samples from different centers and cancer types—lung and head and neck cancer—to test deep features’ predictiveness and two additional datasets to assess the reproducibility of deep features.
Results: Support Vector Machine–Recursive Feature Elimination (SVM–RFE) selected top 100 deep features achieved a concordance index (CI) of 0.67 in survival prediction in LUNG 1, 0.87 in LUNG 4, 0.76 in OPC, and 0.87 in H&N 1, while SVM-RFE selected top 100 radiomics achieved CIs of 0.64, 0.77, 0.73, and 0.74, respectively, all statistically significant differences (p < 0.01, Wilcoxon's test). Most selected deep features are not correlated with tumor volume and TNM staging. However, full radiomics features show higher reproducibility than full deep features in a test/retest setting (0.89 vs. 0.62, concordance correlation coefficient).
Conclusion: The results show that deep features can outperform radiomics while providing different views for tumor prognosis compared to tumor volume and TNM staging. However, deep features suffer from lower reproducibility than radiomic features and lack the interpretability of the latter.