Risk stratification (characterization) of tumors from radiology images can be more accurate and faster with computeraided diagnosis (CAD) tools. Tumor characterization through such tools can also enable non-invasive cancer staging, prognosis, and foster personalized treatment planning as a part of precision medicine. In this study, we propose both supervised and unsupervised machine learning strategies to improve tumor characterization. Our first approach is based on supervised learning for which we demonstrate significant gains with deep learning algorithms, particularly by utilizing a 3D Convolutional Neural Network and Transfer Learning. Motivated by the radiologists' interpretations of the scans, we then show how to incorporate task dependent feature representations into a CAD system via a graph-regularized sparse Multi-Task Learning (MTL) framework.In the second approach, we explore an unsupervised learning algorithm to address the limited availability of labeled training data, a common problem in medical imaging applications. Inspired by learning from label proportion (LLP) approaches in computer vision, we propose to use proportion-SVM for characterizing tumors. We also seek the answer to the fundamental question about the goodness of "deep features" for unsupervised tumor classification. We evaluate our proposed supervised and unsupervised learning algorithms on two different tumor diagnosis challenges: lung and pancreas with 1018 CT and 171 MRI scans, respectively, and obtain the state-of-the-art sensitivity and specificity results in both problems.

Purpose: To evaluate the overall survival of patients with oligometastatic pancreatic ductal adenocarcinoma (PDAC; metastatic tumor <4 cm, ≤2 metastatic tumors total) receiving neoadjuvant therapy, metastasectomy and/or ablation, and primary tumor resection.Methods: We performed a case–control study from January 2005 to December 2015. Patients who underwent curative-intent surgery combined modality therapy (M1 surgery group; 6 [14%], tumor [T]3, node [N]1, and oligo-metastases [M]1) were matched 1 to 3 based on TN stage with two control groups (M0 surgery and M1 no surgery). The M0 surgery group (18 [43%], T3, N1, and M0) included patients without metastases who underwent resection. The M1 no surgery group (18 [43%], T3, N1, and M1) included patients with metastatic PDAC who received palliative chemotherapy without surgical resection.Results: Median overall survival in the M1 surgery, M0 surgery, and M1 no surgery groups was 2.7 years (95% confidence interval [CI], 0.71–3.69), 2.02 years (95% CI, 0.98–3.05), and 0.98 years (95% CI, 0.55–1.25), respectively. Eastern Cooperative Oncology Group (ECOG) status was associated with survival (p = 0.01) after univariate analysis. After adjusting for ECOG status, multivariate analysis showed M1 surgery patients had improved survival compared with M1 no surgery patients and similar survival to M0 surgery patients.Conclusion: Multimodal therapy benefitted our M1 surgery patients. A larger, prospective study of this multidisciplinary management strategy is currently under way.