\n Background\n Patients with neuroendocrine tumours undergoing [\n 177\n Lu]Lu-DOTA-TATE PRRT are at risk of haematological toxicity, and the Osteo-Medullary Invasion Score (OMIS) — a weighted bone-marrow involvement score based on Cristy's somatotope — has been reported to predict this risk but requires labour-intensive manual whole-body contouring. We assessed the technical feasibility of a SwinUNETR-based deep learning pipeline for automated whole-body [\n 68\n Ga]Ga-DOTA-TOC PET/CT lesion segmentation, total metabolic tumour volume (TMTV) quantification, and automated OMIS computation. A retrospective single-centre cohort of 115 patients (188 examinations; mean age 65.5 ± 11.7 years; 52.2% female) was split at the patient level into training (n = 150), validation (n = 19), and test (n = 19) sets. Three architectures were evaluated across four training configurations: Attention U-Net (scratch and FDG-pretrained), nnU-Net, and SwinUNETR. OMIS was computed automatically from predicted lesion masks and TotalSegmentator-derived skeletal anatomy.\nResults\n On the independent test set (n = 19), SwinUNETR achieved a Dice similarity coefficient (DSC) of 0.806 ± 0.123, significantly outperforming the FDG-pretrained Attention U-Net (DSC 0.579 ± 0.189; Wilcoxon p < 0.0001) and comparable to nnU-Net (DSC 0.783 ± 0.181; p = 0.44), with strong biomarker agreement (TMTV r = 0.997, OMIS r = 0.998; both p < 0.0001); for reference, expert inter-observer agreement in this cohort was DSC 0.715. Both OMIS ≥ 30% cases on the test set were correctly classified (2/2 true positives, 17/17 true negatives). Across all partitions including training data (n = 188), sensitivity was 90.2% and specificity 99.3%; the four false negatives all involved diffuse small-volume skeletal disease below the effective detection threshold. The skeletal-attribution component was externally validated on the public ENHANCE.PET 1.6k dataset (whole-skeleton Dice 0.913).\nConclusions\n SwinUNETR demonstrates technical feasibility of automated whole-body lesion segmentation and OMIS computation on [\n 68\n Ga]Ga-DOTA-TOC PET/CT with performance approaching inter-expert agreement. Given the previously reported predictive value of OMIS for haematological toxicity after [\n 177\n Lu]Lu-DOTA-TATE PRRT, this pipeline offers a reproducible approach to automated OMIS quantification. These preliminary results support prospective multicentre validation with haematological outcome data before clinical deployment as a pre-therapeutic risk stratification tool.\n
","journal":null,"year":null,"id":631137,"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":null,"is_data_producer":false,"deposit_databanks":null,"is_oa":false,"file_count":0,"downloads":0,"has_version_chain":false,"published_date":null,"fair_score":null,"fair_percentile":null,"algorithm_id":"datarank_citation_only_1hop_v6","ranking_scope":"data_only","authors":[{"id":1635483,"name":"Solène Malmon","orcid":null,"position":1,"is_corresponding":false},{"id":1635484,"name":"Salim Kanoun","orcid":null,"position":2,"is_corresponding":false},{"id":1635485,"name":"Joris Cocquebert","orcid":null,"position":3,"is_corresponding":false},{"id":1635486,"name":"Thibaut Reichert","orcid":null,"position":4,"is_corresponding":false},{"id":1635487,"name":"Adil Moussali","orcid":null,"position":5,"is_corresponding":false},{"id":1635488,"name":"Nathalie Charrier","orcid":null,"position":6,"is_corresponding":false},{"id":1635489,"name":"Sébastien Benzekry","orcid":null,"position":7,"is_corresponding":false},{"id":1635490,"name":"Daniel Ouk","orcid":null,"position":8,"is_corresponding":false},{"id":1635491,"name":"Sandrine Oziel-Taieb","orcid":null,"position":9,"is_corresponding":false},{"id":1635482,"name":"Romain Ferrara","orcid":"0009-0009-3227-7905","position":0,"is_corresponding":false}],"reference_count":0,"raw_metadata":{"has_enrichment":true,"resolved":true,"title":"Deep Learning-Based Whole-Body Lesion Segmentation and Automated OMIS Computation on [68Ga]Ga-DOTA-TOC PET/CT: Technical Feasibility of a SwinUNETR Pipeline for Pre-PRRT Bone Marrow Involvement Scoring","abstract":"Background Patients with neuroendocrine tumours undergoing [ 177 Lu]Lu-DOTA-TATE PRRT are at risk of haematological toxicity, and the Osteo-Medullary Invasion Score (OMIS) — a weighted bone-marrow involvement score based on Cristy's somatotope — has been reported to predict this risk but requires labour-intensive manual whole-body contouring. We assessed the technical feasibility of a SwinUNETR-based deep learning pipeline for automated whole-body [ 68 Ga]Ga-DOTA-TOC PET/CT lesion segmentation, total metabolic tumour volume (TMTV) quantification, and automated OMIS computation. A retrospective single-centre cohort of 115 patients (188 examinations; mean age 65.5 ± 11.7 years; 52.2% female) was split at the patient level into training (n = 150), validation (n = 19), and test (n = 19) sets. Three architectures were evaluated across four training configurations: Attention U-Net (scratch and FDG-pretrained), nnU-Net, and SwinUNETR. OMIS was computed automatically from predicted lesion masks and TotalSegmentator-derived skeletal anatomy. Results On the independent test set (n = 19), SwinUNETR achieved a Dice similarity coefficient (DSC) of 0.806 ± 0.123, significantly outperforming the FDG-pretrained Attention U-Net (DSC 0.579 ± 0.189; Wilcoxon p < 0.0001) and comparable to nnU-Net (DSC 0.783 ± 0.181; p = 0.44), with strong biomarker agreement (TMTV r = 0.997, OMIS r = 0.998; both p < 0.0001); for reference, expert inter-observer agreement in this cohort was DSC 0.715. Both OMIS ≥ 30% cases on the test set were correctly classified (2/2 true positives, 17/17 true negatives). Across all partitions including training data (n = 188), sensitivity was 90.2% and specificity 99.3%; the four false negatives all involved diffuse small-volume skeletal disease below the effective detection threshold. The skeletal-attribution component was externally validated on the public ENHANCE.PET 1.6k dataset (whole-skeleton Dice 0.913). Conclusions SwinUNETR demonstrates technical feasibility of automated whole-body lesion segmentation and OMIS computation on [ 68 Ga]Ga-DOTA-TOC PET/CT with performance approaching inter-expert agreement. Given the previously reported predictive value of OMIS for haematological toxicity after [ 177 Lu]Lu-DOTA-TATE PRRT, this pipeline offers a reproducible approach to automated OMIS quantification. These preliminary results support prospective multicentre validation with haematological outcome data before clinical deployment as a pre-therapeutic risk stratification tool.
","is_dataset_classified":null,"base_score":0.0,"endowment":0.0,"datacite_reuse_total":0,"file_count":0,"downloads":0,"views":0,"has_version_chain":false,"is_dataset":false,"is_oa":false,"pmid":null,"pmcid":null,"openalex_id":null,"authors":[],"funders":[],"total_grants":0,"fwci":null,"citation_percentile":null,"influential_citations":0,"citation_trend":[],"oa_status":null,"license":null,"oa_locations":[],"fields_of_study":[],"mesh_terms":[],"keywords":[],"sdg_mappings":[],"linked_datasets":[],"clinical_trials":[],"software_tools":[],"database_accessions":[],"source":"live","citation_network_status":"fetched"},"created_at":"2026-08-05T22:54:47.208929Z","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":[]}