BMC artificial intelligence..最新文献

{"title":"End-to-end prediction of clinical outcomes in head and neck squamous cell carcinoma with foundation model-based multiple instance learning.","authors":"Asier Rabasco Meneghetti, Marta Ligero Hernández, Jens-Peter Kühn, Steffen Löck, Zunamys Itzell Carrero, Raquel Perez-Lopez, Keno K Bressem, Titus J Brinker, Alexander T Pearson, Daniel Truhn, Sven Nebelung, Jakob Nikolas Kather","doi":"10.1186/s44398-025-00003-8","DOIUrl":"10.1186/s44398-025-00003-8","url":null,"abstract":"<p><strong>Background: </strong>Foundation models have shown promise in medical AI by learning flexible features from large datasets, offering new opportunities for improving endpoint prediction. However, usage of foundation models for endpoint prediction using routine imaging in head and neck squamous cell carcinoma patients remains unexplored. Within this study, we evaluated the potential of foundation-model based multiple instance learning for prediction of 2-year overall survival, locoregional control and freedom from distant metastasis across three external head and neck squamous cell carcinoma patient cohorts using 2D, multiview and 3D approaches while comparing prediction and stratification performance with handcrafted radiomics and clinical baselines.</p><p><strong>Results: </strong>2D multiple-instance learning models achieved 2-year test area under the receiver-operator curve (AUROC) range of 0.75-0.84 for 2-year overall survival, 0.66-0.75 for 2-year locoregional control and 0.71-0.78 for 2-year freedom from distant metastasis across three different external cohorts, outperforming multiview and 3D multiple instance learning models (AUROC range: 0.50-0.77, p <math><mo>≥</mo></math> 0.15) and showing comparable or superior performance to handcrafted radiomics (AUROC range: 0.64-0.74, p <math><mo>≥</mo></math> 0.012). Significant stratification was observed from the 2D MIL models (hazard ratios: 2.14-4.77, p <math><mo>≤</mo></math> 0.039). 2D MIL models were also shown to learn endpoint-specific correlation patterns such as N-stage for 2-year freedom from distant metastasis prognosis. Multimodal enhancement of 2-year OS/FFDM (AUROC range: 0.82-0.87, p <math><mo>≤</mo></math> 0.018) for patients without human papilloma virus positive tumors.</p><p><strong>Conclusions: </strong>FM-based 2D MIL demonstrates promise in HNSCC risk prediction as well as stratification of clinical outcomes. The models match or outperform radiomics baselines, learning clinically-related patterns and showing enhancement of clinical baselines in non-human papilloma virus positive patients.</p><p><strong>Supplementary information: </strong>The online version contains supplementary material available at 10.1186/s44398-025-00003-8.</p>","PeriodicalId":520917,"journal":{"name":"BMC artificial intelligence..","volume":"1 1","pages":"3"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12212421/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144556570","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Hyperspectral reconstruction for mobile diabetic foot blood perfusion monitoring.","authors":"Alexander Gherardi, Wei Bo, Ahmet Demirbas, Ye Zhan, Wenyao Xu","doi":"10.1186/s44398-025-00011-8","DOIUrl":"10.1186/s44398-025-00011-8","url":null,"abstract":"<p><strong>Background: </strong>Blood Perfusion is a key factor in the development and healing of wounded tissues including Diabetic Foot Ulcers (DFU), a harmful chronic wound caused by diabetic neuropathy. Recent works have explored the use of hyperspectral imaging (HSI) to non-invasively quantify the quality of blood perfusion with high spatial resolution. Later works consider the use of hyperspectral reconstruction (HSR) to provide the same capability using unmodified commodity hardware, such as smartphone cameras, using computational methods to yield full hyperspectral images from RGB ones. However, these HSR perfusion systems require profiles for each camera they are used with and furthermore require radiometric calibration to account for environmental lighting conditions before each use.</p><p><strong>Methods: </strong>In this work we demonstrate MobiPerf which extracts oxygenation signals/images along with high fidelity remote PPG signals while overcoming these challenges. To eliminate the need for camera profiles, our system uses deep learning HSR models that have been shown to generalize well across different cameras. Then to overcome the need for reference image calibration, we utilize a custom algorithm <i>Calibration Free Skin Compensation Estimation</i>.</p><p><strong>Results: </strong>Evaluated under 5 different simulated lighting conditions from the CIE Standard Illuminates, our system maintains strong agreement with oxygenation images/signals extracted directly from HSI cameras. Our testing on in-the-wild RGB data from a publicly available dataset of diabetic foot ulcer images (N [Formula: see text] 6000) shows an acute sensitivity to Ischemia conditions (p [Formula: see text]) as well as a more limited sensitivity to infection complications. Along with a dataset of videos with contract PPG (N = 56) which shows rPPG performance on par or better than other state-of-the-art algorithms.</p><p><strong>Conclusions: </strong>Our results demonstrate that a HSR system can be used to monitor diabetic foot ulcers using just images/videos minimizing the need for procedures prior to or during use and with mobile hardware patients already have. We anticipate that in the future our advancements in HSR can be used for other smart health applications that relate to perfusion, and we anticipate that similar HSR based systems can be used to monitor other tissue parameters such as sweat concentrations.</p>","PeriodicalId":520917,"journal":{"name":"BMC artificial intelligence..","volume":"1 1","pages":"10"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12454483/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145139924","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}