{"doi":"10.1145/3765612.3767232","title":"Longitudinal Tumor Generation in Mammograms via Dual Encoder GAN and Learnable Blending","abstract":"Despite progress in deep learning for breast cancer detection, model development is still limited by the lack of annotated mammograms, particularly those that capture tumor development over time. Most existing methods rely solely on current images and focus on reconstructing known patterns, without modeling how tumors might develop at different timepoints. Generating realistic tumors in highresolution full-field digital mammograms (FFDM) is also challenging due to complex textures and subtle structures. To address these gaps, we propose a new end-to-end generative framework that creates synthetic tumor images based on prior and current normal mammograms, effectively simulating longitudinal cancer cases. Our model uses dual Transformer-based encoders to extract features from each timepoint and a variational latent space to generate diverse tumor appearances. We also present a Transformer-based decoder that fuses latent and anatomical features through attention mechanisms for context-aware tumor reconstruction. A differentiable soft-mask blending module adaptively inserts the synthesized tumor into the current mammogram. To support both supervised and weakly supervised training, we design a hybrid loss combining reconstruction, KL divergence, tumor, and adversarial components. Experimental results using FID, FMD, radiomics, and classification metrics demonstrate that our model generates synthetic tumors closely resembling real ones, supporting realistic synthesis, temporal modeling, and data augmentation in longitudinal mammography. The code is available at: https://github.com/NabaviLab/SynGAN.","journal":null,"year":2025,"id":586267,"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":0,"citer_count":0,"citers_with_citation_signal":0,"citers_with_endowment":0,"datacite_reuse_total":0,"is_dataset":false,"is_dataset_confidence":0.9339,"is_data_producer":false,"deposit_databanks":null,"is_oa":true,"file_count":0,"downloads":0,"has_version_chain":false,"published_date":"2025-01-01","fair_score":null,"fair_percentile":null,"algorithm_id":"datarank_citation_only_1hop_v6","ranking_scope":"data_only","authors":[{"id":1166233,"name":"Sahand Hamzehei","orcid":"0009-0002-2015-9820","position":1,"is_corresponding":false},{"id":1338875,"name":"Mostafa Karami","orcid":"0009-0009-1083-2138","position":2,"is_corresponding":false},{"id":1500287,"name":"Stephen Andrew Baker","orcid":"0009-0003-8866-2999","position":3,"is_corresponding":false},{"id":1500288,"name":"Tucker Van Rathe","orcid":"0009-0005-3986-1582","position":4,"is_corresponding":false},{"id":1331970,"name":"Clifford Yang","orcid":"0000-0001-7108-2021","position":5,"is_corresponding":false},{"id":446112,"name":"Sheida Nabavi","orcid":"0000-0002-5996-1020","position":6,"is_corresponding":false},{"id":1331968,"name":"Afsana Ahsan Jeny","orcid":"0000-0002-8524-9600","position":0,"is_corresponding":true}],"reference_count":23,"raw_metadata":null,"created_at":"2026-07-19T02:59:28.666390Z","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":[]}