{"doi":"10.1016/j.ostima.2021.100003","title":"An ensemble clinical and MR-image deep learning model predicts 8-year knee pain trajectory: Data from the osteoarthritis initiative","abstract":"We applied data-driven feature learning to classify the presence and 8-years incidence of knee pain from MRI. We analyzed the data from the Osteoarthritis Initiative. Self-reported pain scores from knees with chronic pain were used to generate binary labels for the pain presence. A 3D DenseNet was then trained to classify the presence of pain from sagittal intermediate-weighted 2D turbo spin-echo fat suppression MRI. Next, functional principal component analysis (FPCA) was performed to model the temporal patterns of pain incidence among non-symptomatic knees at baseline. Bayesian Gaussian mixture models was fitted to the FPCA scores to identify clusters of pain trajectories. Cluster membership were used as labels and learned by the 3D DenseNet from MRI only and MRI combined with clinical features. The pain presence and pain incidence prediction models included 14,606 and 3,828 samples, respectively. The pain presence classification model achieved an AUC of 0.898 (95%CI: 0.882-0.914). We identified two distinct subtypes of the temporal progression of pain score: stable (87%) and worsening group (13%). Our model showed high performance in subtype classification using both imaging and clinical features (AUC: 0.794 [95%CI: 0.761-0.823]). Our data-driven approach showed high performance in extracting MRI features associated with chronic knee pain presence and incident knee pain. Our results provide a basis for a more comprehensive quantitative analysis of MRI to uncover latent relationships between OA imaging and knee pain. Further studies are needed to improve our understanding of the extracted features.","journal":"Osteoarthritis Imaging","year":2021,"id":192411,"datarank":0.0,"base_score":0.0,"endowment":0.0,"self_citation_contribution":0.0,"citation_network_contribution":0.0,"self_endowment_contribution":0.0,"citer_contribution":0.0,"corpus_percentile":null,"corpus_rank":null,"citation_count":12,"citer_count":0,"citers_with_citation_signal":0,"citers_with_endowment":0,"datacite_reuse_total":0,"is_dataset":false,"is_dataset_confidence":0.9177,"is_data_producer":false,"deposit_databanks":null,"is_oa":true,"file_count":0,"downloads":0,"has_version_chain":false,"published_date":"2021-01-01","fair_score":null,"fair_percentile":null,"algorithm_id":"datarank_citation_only_1hop_v6","ranking_scope":"data_only","authors":[{"id":593702,"name":"Felix Liu","orcid":"0000-0001-8288-2817","position":1,"is_corresponding":false},{"id":32193,"name":"Sharmila Majumdar","orcid":"0000-0002-0201-871X","position":2,"is_corresponding":false},{"id":22114,"name":"Valentina Pedoia","orcid":"0000-0002-9745-955X","position":3,"is_corresponding":false},{"id":342366,"name":"Jinhee J. Lee","orcid":"0000-0002-1411-5496","position":0,"is_corresponding":true}],"reference_count":41,"raw_metadata":null,"created_at":"2026-07-18T23:49:43.496702Z","pmid":null,"pmcid":null,"fwci":null,"citation_percentile":null,"influential_citations":0,"oa_status":null,"license":null,"views":0,"total_file_size_bytes":0,"version_count":0,"fair_f":null,"fair_a":null,"fair_i":null,"fair_r":null,"fair_zscore":null,"fair_rationale":null,"fair_model":null,"fair_agent_version":null,"fair_fulltext_source":null,"fair_has_llm":null,"fair_computed_at":null,"clinical_trials":[],"software_tools":[],"db_accessions":[],"linked_datasets":[],"topics":[]}