Martin Lillholm

Prof. Martin Lillholm

orcid

Research Focus & Metrics

Academic Career Length

7Years (2020-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 July 2022 in Radiology
Verified
Journal Article
Closed access
ALAndreas D. Lauritzen
Andreas D. Lauritzen
ARAlejandro Rodríguez-Ruiz
Alejandro Rodríguez-Ruiz
MEMy Catarina von Euler-Chelpin
My Catarina von Euler-Chelpin
ELElsebeth Lynge
Elsebeth Lynge
IVIlse Vejborg
Ilse Vejborg
MLMartin Lillholm
Martin Lillholm
In a simulation study of breast cancer screenings in more than 114 000 women, an artificial intelligence–based mammography screening protocol had similar sensitivity to that of radiologist screenin... Background Developments in artificial intelligence (AI) systems to assist radiologists in reading mammograms could improve breast cancer screening efficiency. Purpose To investigate whether an AI s...

1 September 2023 in Journal of Medical Imaging
Verified
Journal Article
Closed access
PDF file
ALAndreas D. Lauritzen
Andreas D. Lauritzen
MEMy Catarina von Euler-Chelpin
My Catarina von Euler-Chelpin
ELElsebeth Lynge
Elsebeth Lynge
IVIlse Vejborg
Ilse Vejborg
MNMads Nielsen
Mads Nielsen
MLMartin Lillholm
Martin Lillholm
PurposeRisk-stratified breast cancer screening might improve early detection and efficiency without comprising quality. However, modern mammography-based risk models do not ensure adaptation across vendor-domains and rely on cancer precursors, associated with short-term risk, which might limit long-term risk assessment. We report a cross-vendor mammographic texture model for long-term risk.ApproachThe texture model was robustly trained using two systematically designed case-control datasets. Textural features, indicative of future breast cancer, were learned by excluding samples with diagnosed/potential malignancies from training. An augmentation-based domain adaption technique, based on flavorization of mammographic views, ensured generalization across vendor-domains. The model was validated in 66,607 consecutively screened Danish women with flavorized Siemens views and 25,706 Dutch women with Hologic-processed views. Performances were evaluated for interval cancers (IC) within 2 years from screening and long-term cancers (LTC) from 2 years after screening. The texture model was combined with established risk factors to flag 10% of women with the highest risk.ResultsIn Danish women, the texture model achieved an area under the receiver operating characteristic curve (AUC) of 0.71 and 0.65 for ICs and LTCs, respectively. In Dutch women with Hologic-processed views, the AUCs were not different from AUCs in Danish women with flavorized views. The AUC for texture combined with established risk factors increased to 0.68 for LTCs. The 10% of women flagged as high-risk accounted for 25.5% of ICs and 24.8% of LTCs.ConclusionsThe texture model robustly estimated long-term breast cancer risk while adapting to an unseen processed vendor-domain and identified a clinically relevant high-risk subgroup.

21 August 2024 in Diagnostics
Verified
Journal Article
Open access
PDF file
BVBolette Mikela Vilmun
Bolette Mikela Vilmun
GNGeorge Napolitano
George Napolitano
ALAndreas Lauritzen
Andreas Lauritzen
ELElsebeth Lynge
Elsebeth Lynge
MLMartin Lillholm
Martin Lillholm
IVIlse Vejborg
Ilse Vejborg
Assessing a woman’s risk of breast cancer is important for personalized screening. Mammographic density is a strong risk factor for breast cancer, but parenchymal texture patterns offer additional information which cannot be captured by density. We aimed to combine BI-RADS density score 4th Edition and a deep-learning-based texture score to stratify women in screening and compare rates among the combinations. This retrospective study cohort study included 216,564 women from a Danish populations-based screening program. Baseline mammograms were evaluated using BI-RADS density scores (1–4) and a deep-learning texture risk model, with scores categorized into four quartiles (1–4). The incidence rate ratio (IRR) for screen-detected, interval, and long-term cancer were adjusted for age, year of screening and screening clinic. Compared with subgroup B1-T1, the highest IRR for screen-detected cancer were within the T4 category (3.44 (95% CI: 2.43–4.82)−4.57 (95% CI: 3.66–5.76)). IRR for interval cancer was highest in the BI-RADS 4 category (95% CI: 5.36 (1.77–13.45)−16.94 (95% CI: 9.93–30.15)). IRR for long-term cancer increased both with increasing BI-RADS and increasing texture reaching 5.15 (4.31–6.16) for the combination of B4-T4 compared with B1-T1. Deep-learning-based texture analysis combined with BI-RADS density categories can reveal subgroups with increased rates beyond what density alone can ascertain, suggesting the potential of combining texture and density to improve risk stratification in breast cancer screening.

11 October 2020 in medRxiv
Verified
Preprint
Open access
PDF file
EJEspen Jimenez-Solem
Espen Jimenez-Solem
TPTonny Studsgaard Petersen
Tonny Studsgaard Petersen
CHCasper Hansen
Casper Hansen
CHChristian Hansen
Christian Hansen
CLChristina Lioma
Christina Lioma
MSMartin Sillesen
Martin Sillesen
Background: Patients with severe COVID-19 have overwhelmed healthcare systems worldwide. We hypothesized that Machine Learning (ML) models could be used to predict risks at different stages of management (at diagnosis, hospital admission and ICU admission) and thereby provide insights into drivers and prognostic markers of disease progression and death. Methods: From a cohort of approx. 2.6 million citizens in the two regions of Denmark, SARS-CoV-2 PCR tests were performed on subjects suspected for COVID-19 disease; 3944 cases had at least one positive test and were subjected to further analysis. A cohort of SARS- CoV-2 positive cases from the United Kingdom Biobank was used for external validation. Findings: The ML models predicted the risk of death (Receiver Operation Characteristics – Area Under the Curve, ROC-AUC) of 0.904 at diagnosis, 0.818, at hospital admission and 0.723 at Intensive Care Unit (ICU) admission. Similar metrics were achieved for predicted risks of hospital and ICU admission and use of mechanical ventilation. We identified some common risk factors, including age, body mass index (BMI) and hypertension as driving factors, although the top risk features shifted towards markers of shock and organ dysfunction in ICU patients. The external validation indicated fair predictive performance for mortality prediction, but suboptimal performance for predicting ICU admission. Interpretation: ML may be used to identify drivers of progression to more severe disease and for prognostication patients in patients with COVID-19. Prognostic features included age, BMI and hypertension, although markers of shock and organ dysfunction became more important in more severe cases. We provide access to an online risk calculator based on these findings. Funding: The study was funded by grants from the Novo Nordisk Foundation to MS (#NNF20SA0062879 and #NNF19OC0055183) and MN (#NNF20SA0062879). The foundation took no part in project design, data handling and manuscript preparation.

16 October 2026 in Journal of Human-Technology Relations
Verified
Journal Article
Open access
PDF file
MNMaria Bach Nielsen
Maria Bach Nielsen
Tom BørsenTom Børsen
Tom Børsen
MLMartin Lillholm
Martin Lillholm
ACAlessandra Cenci
Alessandra Cenci
Artificial intelligence is increasingly deployed in healthcare. Radiology is one of the most active areas of clinical AI development. Despite substantial investments, many AI systems struggle to gain traction in clinical practice. The issues are rarely purely technical; they concern how AI tools fit into existing workflows, how they interact with professional responsibilities, and how they sit within institutional arrangements. It also remains unclear how core values in AI research such as explainability, value alignment, accountability, and trust are negotiated in clinical work. Established Health Technology Assessment (HTA) models define assessment domains but offer limited guidance on implementation: how technologies are staged within workflows, how responsibilities are redistributed, and how local conditions redefine value for different stakeholders. This paper presents the AI in PRACTICE framework, a five-phase methodology grounded in Techno-Anthropology and its relational ontology, for examining the situated work through which AI becomes part of clinical practice. Explainability, value alignment, trust, and distributed responsibility are understood as produced through socio-technical practices. The framework complements existing HTA models through iterative engagement and stakeholder participation. Insights are illustrated through three radiology-facing AI projects showing how assessment criteria are enacted, resisted, and revised in context, and how the framework generates practice-based artefacts.

30 July 2026 in Communications Medicine
Verified
Journal Article
Open access
PDF file
SISilvia Ingala
Silvia Ingala
MBMarko Bauer
Marko Bauer
IWIvar I. J. H. G. Wamelink
Ivar I. J. H. G. Wamelink
MPMathias Perslev
Mathias Perslev
KSKaining Sheng
Kaining Sheng
APAkshay Pai
Akshay Pai
Background: Gadolinium-based contrast agents used for brain magnetic resonance imaging improve lesion detection but are associated with acute side effects, concerns about long-term tissue deposition, and economic and environmental costs. Regulatory bodies and radiological societies therefore recommend minimizing their use where possible. This study aims to develop and evaluate a real-time artificial intelligence algorithm, Apollo SmartGAD, that optimizes imaging workflows by recommending contrast-enhanced, reduced-dose contrast-enhanced, or non-enhanced imaging protocols during acquisition. Methods: This retrospective, multicenter study evaluated 1251 adult magnetic resonance imaging examinations (mean age 52.1 ± 20.4 years; 577 females). Apollo SmartGAD analyzes images acquired during scanning and recommends non-enhanced, reduced-dose contrast-enhanced, or standard-dose contrast-enhanced imaging protocols. Performance was assessed in two distinct cohorts: patients undergoing first-time brain assessment and patients undergoing glioma monitoring. Results: Here we show that, in the first-time brain assessment cohort, Apollo SmartGAD identifies examinations not requiring contrast administration with a sensitivity of 73%, a specificity of 79%, and a negative predictive value of 86%. In the glioma monitoring cohort, the algorithm differentiate between non-enhanced, reduced-dose contrast-enhanced, and standard-dose contrast-enhanced protocols with an overall sensitivity of 77% and a specificity of 78%. Conclusions: This study demonstrates that workflow-integrated, real-time artificial intelligence can support contrast administration decisions during brain magnetic resonance imaging. Such decision support may reduce unnecessary contrast use while preserving diagnostic performance, with potential benefits for patient safety, clinical efficiency, and sustainability.