Ole Emil Andersen
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Ole Emil Andersen has a total of 43 co-authors. Here are the top co-authors by number of shared publications:
- Kristian Overgaard
Department of Public Health, Section of Sport Science Aarhus University Aarhus Denmark
4 shared publications - Adam Hulman
Department of Public Health, Aarhus University, Aarhus, Denmark
2 shared publications - Andreas Ebbehoj
Division of Endocrinology, Diabetes, Metabolism and Nutrition, Mayo Clinic , Rochester, Minnesota
2 shared publications - Jeppe Foged Vigh-Larsen
Department of Sports Science and Clinical Biomechanics, SDU Sport and Health Sciences Cluster (SHSC) University of Southern Denmark Odense Denmark
2 shared publications - Joachim Nielsen
Department of Sports Science and Clinical Biomechanics University of Southern Denmark Odense Denmark
2 shared publications - Mette Østergaard Thunbo
Department of Clinical Medicine, Aarhus University, Denmark
2 shared publications
Latest Publications
- 5 February 2026
- Published by Wiley in Acta Physiologica
- Vol. 242 (3) , e70161
Ketone Monoester Increases Skeletal Muscle Power and Energy Turnover in Older but Not Young Men Without Affecting Metabolic Economy: A Controlled, Double Blind, Cross‐Over Trial
Aim: Ketosis may represent a therapeutic target for age‐related impairments in skeletal muscle function. This study investigated acute effects of ketosis on metabolic economy, mitochondrial function, and contractile parameters in skeletal muscle of young and older adults. Methods: Twelve young (20–25 years) and twelve older (65–85 years) healthy men, matched by age‐adjusted V̇O 2 max, participated in a randomized, crossover, double‐blind intervention with ingestion of ketone monoester or placebo on separate study days. On both days, a low‐dose, continuous glucose infusion blocked endogenous ketone production. Metabolic economy, oxidative capacity, muscle performance, intramuscular pH, and relative decline in peak power were assessed in the tibialis anterior through phosphorous MR spectroscopy ( 31 P‐MRS) and dynamometer recordings. Mitochondrial function of the quadriceps femoris muscle was assessed by high‐resolution respirometry. Results: Ketosis had no effect on metabolic economy in either young or older participants. The older group showed lower metabolic economy compared to the young group. In older participants, ketones increased ATP production and time‐torque derived work capacity. Oxidative capacity was similar between groups and remained unaffected by ketones. In the older group, ketones improved peak power and increased both muscle relative decline in peak power and contraction‐induced pH decline. Complex I + II respiration was lower in older compared to young participants, with no effect of ketones. Conclusion: Ketosis enhanced skeletal muscle work capacity and ATP production in older but not young adults, suggesting an age‐specific effect of ketone bodies on muscle function that operates independently of changes in metabolic economy and mitochondrial function. These findings support ketosis as a promising ergogenic therapy for older adults. Trial Registration: The study was pre‐registered at clinicaltrials.gov (NCT05732909)
- 5 February 2026
- Published by Wiley in Acta Physiologica
- Vol. 242 (3) , e70161
Ketone Monoester Increases Skeletal Muscle Power and Energy Turnover in Older but Not Young Men Without Affecting Metabolic Economy: A Controlled, Double Blind, Cross‐Over Trial
Aim: Ketosis may represent a therapeutic target for age‐related impairments in skeletal muscle function. This study investigated acute effects of ketosis on metabolic economy, mitochondrial function, and contractile parameters in skeletal muscle of young and older adults. Methods: Twelve young (20–25 years) and twelve older (65–85 years) healthy men, matched by age‐adjusted V̇O 2 max, participated in a randomized, crossover, double‐blind intervention with ingestion of ketone monoester or placebo on separate study days. On both days, a low‐dose, continuous glucose infusion blocked endogenous ketone production. Metabolic economy, oxidative capacity, muscle performance, intramuscular pH, and relative decline in peak power were assessed in the tibialis anterior through phosphorous MR spectroscopy ( 31 P‐MRS) and dynamometer recordings. Mitochondrial function of the quadriceps femoris muscle was assessed by high‐resolution respirometry. Results: Ketosis had no effect on metabolic economy in either young or older participants. The older group showed lower metabolic economy compared to the young group. In older participants, ketones increased ATP production and time‐torque derived work capacity. Oxidative capacity was similar between groups and remained unaffected by ketones. In the older group, ketones improved peak power and increased both muscle relative decline in peak power and contraction‐induced pH decline. Complex I + II respiration was lower in older compared to young participants, with no effect of ketones. Conclusion: Ketosis enhanced skeletal muscle work capacity and ATP production in older but not young adults, suggesting an age‐specific effect of ketone bodies on muscle function that operates independently of changes in metabolic economy and mitochondrial function. These findings support ketosis as a promising ergogenic therapy for older adults. Trial Registration: The study was pre‐registered at clinicaltrials.gov (NCT05732909)
- Open Access
- 22 September 2024
- Published by Wiley in Scandinavian Journal of Medicine & Science in Sports
- Vol. 34 (9) , e14731
No Effects of Carbohydrate Ingestion on Muscle Metabolism or Performance During Short‐Duration High‐Intensity Intermittent Exercise
Carbohydrates are critical for high‐intensity exercise performance. However, the effects of carbohydrate supplementation on muscle metabolism and performance during short‐duration high‐intensity intermittent exercise remain inadequately explored. Our aim was to address this aspect in a randomized, counterbalanced, double‐blinded crossover design. Eleven moderately‐to‐well‐trained males performed high‐intensity intermittent cycling receiving carbohydrate (CHO, ~55 g/h) or placebo (PLA) fluid supplementation. Three exercise periods (EX1‐EX3) were completed comprising 10 × 45 s at ~105% Wmax interspersed with 135 s rest between bouts and ~20 min between periods. Repeated sprint ability (5 × 6 s sprints with 24 s recovery) was assessed at baseline and after each period. Thigh muscle biopsies were obtained at baseline and before and after EX3 to determine whole‐muscle and fiber‐type‐specific glycogen depletion. No differences were found in muscle glycogen degradation at the whole‐muscle (p = 0.683) or fiber‐type‐specific level (p = 0.763–0.854) with similar post‐exercise whole‐muscle glycogen concentrations (146 ± 20 and 122 ± 15 mmol·kg−1 dw in CHO and PLA, respectively). Repeated sprint ability declined by ~9% after EX3 with no between‐condition differences (p = 0.971) and no overall differences in ratings of perceived exertion (p = 0.550). This was despite distinctions in blood glucose concentrations throughout exercise, reaching post‐exercise levels of 5.3 ± 0.2 and 4.1 ± 0.2 mmol·L−1 (p < 0.001) in CHO and PLA, respectively, accompanied by fivefold higher plasma insulin levels in CHO (p < 0.001). In conclusion, we observed no effects of carbohydrate ingestion on net muscle glycogen breakdown or sprint performance during short‐duration high‐intensity intermittent exercise despite elevated blood glucose and insulin levels. These results therefore question the efficacy of carbohydrate supplementation strategies in high‐intensity intermittent sports.
- 22 September 2024
- Published by Wiley in Scandinavian Journal of Medicine & Science in Sports
- Vol. 34 (9) , e14731
Open AccessNo Effects of Carbohydrate Ingestion on Muscle Metabolism or Performance During Short‐Duration High‐Intensity Intermittent Exercise
Carbohydrates are critical for high‐intensity exercise performance. However, the effects of carbohydrate supplementation on muscle metabolism and performance during short‐duration high‐intensity intermittent exercise remain inadequately explored. Our aim was to address this aspect in a randomized, counterbalanced, double‐blinded crossover design. Eleven moderately‐to‐well‐trained males performed high‐intensity intermittent cycling receiving carbohydrate (CHO, ~55 g/h) or placebo (PLA) fluid supplementation. Three exercise periods (EX1‐EX3) were completed comprising 10 × 45 s at ~105% Wmax interspersed with 135 s rest between bouts and ~20 min between periods. Repeated sprint ability (5 × 6 s sprints with 24 s recovery) was assessed at baseline and after each period. Thigh muscle biopsies were obtained at baseline and before and after EX3 to determine whole‐muscle and fiber‐type‐specific glycogen depletion. No differences were found in muscle glycogen degradation at the whole‐muscle (p = 0.683) or fiber‐type‐specific level (p = 0.763–0.854) with similar post‐exercise whole‐muscle glycogen concentrations (146 ± 20 and 122 ± 15 mmol·kg−1 dw in CHO and PLA, respectively). Repeated sprint ability declined by ~9% after EX3 with no between‐condition differences (p = 0.971) and no overall differences in ratings of perceived exertion (p = 0.550). This was despite distinctions in blood glucose concentrations throughout exercise, reaching post‐exercise levels of 5.3 ± 0.2 and 4.1 ± 0.2 mmol·L−1 (p < 0.001) in CHO and PLA, respectively, accompanied by fivefold higher plasma insulin levels in CHO (p < 0.001). In conclusion, we observed no effects of carbohydrate ingestion on net muscle glycogen breakdown or sprint performance during short‐duration high‐intensity intermittent exercise despite elevated blood glucose and insulin levels. These results therefore question the efficacy of carbohydrate supplementation strategies in high‐intensity intermittent sports.
- Open Access
- 11 July 2023
- Published by MDPI AG in Diagnostics
- Vol. 13 (14) , 2345
Usefulness of Heat Map Explanations for Deep-Learning-Based Electrocardiogram Analysis
Deep neural networks are complex machine learning models that have shown promising results in analyzing high-dimensional data such as those collected from medical examinations. Such models have the potential to provide fast and accurate medical diagnoses. However, the high complexity makes deep neural networks and their predictions difficult to understand. Providing model explanations can be a way of increasing the understanding of “black box” models and building trust. In this work, we applied transfer learning to develop a deep neural network to predict sex from electrocardiograms. Using the visual explanation method Grad-CAM, heat maps were generated from the model in order to understand how it makes predictions. To evaluate the usefulness of the heat maps and determine if the heat maps identified electrocardiogram features that could be recognized to discriminate sex, medical doctors provided feedback. Based on the feedback, we concluded that, in our setting, this mode of explainable artificial intelligence does not provide meaningful information to medical doctors and is not useful in the clinic. Our results indicate that improved explanation techniques that are tailored to medical data should be developed before deep neural networks can be applied in the clinic for diagnostic purposes.
- 11 July 2023
- Published by MDPI AG in Diagnostics
- Vol. 13 (14) , 2345
Open AccessUsefulness of Heat Map Explanations for Deep-Learning-Based Electrocardiogram Analysis
Deep neural networks are complex machine learning models that have shown promising results in analyzing high-dimensional data such as those collected from medical examinations. Such models have the potential to provide fast and accurate medical diagnoses. However, the high complexity makes deep neural networks and their predictions difficult to understand. Providing model explanations can be a way of increasing the understanding of “black box” models and building trust. In this work, we applied transfer learning to develop a deep neural network to predict sex from electrocardiograms. Using the visual explanation method Grad-CAM, heat maps were generated from the model in order to understand how it makes predictions. To evaluate the usefulness of the heat maps and determine if the heat maps identified electrocardiogram features that could be recognized to discriminate sex, medical doctors provided feedback. Based on the feedback, we concluded that, in our setting, this mode of explainable artificial intelligence does not provide meaningful information to medical doctors and is not useful in the clinic. Our results indicate that improved explanation techniques that are tailored to medical data should be developed before deep neural networks can be applied in the clinic for diagnostic purposes.
- Open Access
- 27 March 2023
- Published by Wiley in Scandinavian Journal of Medicine & Science in Sports
- Vol. 33 (7) , 1054-1071
Lowered muscle glycogen reduces body mass with no effect on short‐term exercise performance in men
Performance in short‐duration sports is highly dependent on muscle glycogen, but the total degradation is only moderate and considering the water‐binding property of glycogen, unnecessary storing of glycogen may cause an unfavorable increase in body mass. To investigate this, we determined the effect of manipulating dietary carbohydrates (CHO) on muscle glycogen content, body mass, and short‐term exercise performance. In a randomized and counterbalanced cross‐over design, twenty‐two men completed two maximal cycle tests of either 1‐min (n = 10) or 15‐min (n = 12) duration with different pre‐exercise muscle glycogen levels. Glycogen manipulation was initiated three days prior to the tests by exercise‐induced glycogen depletion followed by ingestion of a moderate (M‐CHO) or high (H‐CHO) CHO‐diet. Subjects were weighed before each test, and muscle glycogen content was determined in biopsies from m. vastus lateralis before and after each test. Pre‐exercise muscle glycogen content was lower following M‐CHO than H‐CHO (367 mmol · kg−1 DW vs. 525 mmol · kg−1 DW, p < 0.00001), accompanied by a 0.7 kg lower body mass (p < 0.00001). No differences were observed in performance between diets in neither the 1‐min (p = 0.33) nor the 15‐min (p = 0.99) test. In conclusion, pre‐exercise muscle glycogen content and body mass were lower after ingesting moderate compared with high amounts of CHO, while short‐term exercise performance was unaffected. This demonstrates that adjusting pre‐exercise glycogen levels to the requirements of competition may provide an attractive weight management strategy in weight‐bearing sports, particularly in athletes with high resting glycogen levels.
- 27 March 2023
- Published by Wiley in Scandinavian Journal of Medicine & Science in Sports
- Vol. 33 (7) , 1054-1071
Open AccessLowered muscle glycogen reduces body mass with no effect on short‐term exercise performance in men
Performance in short‐duration sports is highly dependent on muscle glycogen, but the total degradation is only moderate and considering the water‐binding property of glycogen, unnecessary storing of glycogen may cause an unfavorable increase in body mass. To investigate this, we determined the effect of manipulating dietary carbohydrates (CHO) on muscle glycogen content, body mass, and short‐term exercise performance. In a randomized and counterbalanced cross‐over design, twenty‐two men completed two maximal cycle tests of either 1‐min (n = 10) or 15‐min (n = 12) duration with different pre‐exercise muscle glycogen levels. Glycogen manipulation was initiated three days prior to the tests by exercise‐induced glycogen depletion followed by ingestion of a moderate (M‐CHO) or high (H‐CHO) CHO‐diet. Subjects were weighed before each test, and muscle glycogen content was determined in biopsies from m. vastus lateralis before and after each test. Pre‐exercise muscle glycogen content was lower following M‐CHO than H‐CHO (367 mmol · kg−1 DW vs. 525 mmol · kg−1 DW, p < 0.00001), accompanied by a 0.7 kg lower body mass (p < 0.00001). No differences were observed in performance between diets in neither the 1‐min (p = 0.33) nor the 15‐min (p = 0.99) test. In conclusion, pre‐exercise muscle glycogen content and body mass were lower after ingesting moderate compared with high amounts of CHO, while short‐term exercise performance was unaffected. This demonstrates that adjusting pre‐exercise glycogen levels to the requirements of competition may provide an attractive weight management strategy in weight‐bearing sports, particularly in athletes with high resting glycogen levels.
- 1 March 2023
- Published by American Physiological Society in Journal of Applied Physiology
- Vol. 134 (3) , 777-785
Force potentiation during eccentric contractions in rat skeletal muscle
Force during eccentric contractions can increase to a magnitude that may have profound consequences for our understanding of skeletal muscle locomotion. This increase in eccentric force occurs over consecutive, shortly interspaced, tetanic contractions in rat extensor digitorum longus muscles—not in rat soleus muscles—and coincides with well-known traits of postactivation potentiation. Eccentric force potentiation may significantly enhance muscle performance in activities involving stretch-shortening cycles.
- 1 March 2023
- Published by American Physiological Society in Journal of Applied Physiology
- Vol. 134 (3) , 777-785
Force potentiation during eccentric contractions in rat skeletal muscle
Force during eccentric contractions can increase to a magnitude that may have profound consequences for our understanding of skeletal muscle locomotion. This increase in eccentric force occurs over consecutive, shortly interspaced, tetanic contractions in rat extensor digitorum longus muscles—not in rat soleus muscles—and coincides with well-known traits of postactivation potentiation. Eccentric force potentiation may significantly enhance muscle performance in activities involving stretch-shortening cycles.
- Open Access
- 18 September 2022
- Published by Wiley in The Journal of Physiology
- Vol. 600 (21) , 4713-4730
Fibre type‐ and localisation‐specific muscle glycogen utilisation during repeated high‐intensity intermittent exercise
Glycogen particles are situated in key areas of the muscle cell in the vicinity of the main energy‐consumption sites and may be utilised heterogeneously dependent on the nature of the metabolic demands. The present study aimed to investigate the time course of fibre type‐specific utilisation of muscle glycogen in three distinct subcellular fractions (intermyofibrillar, IMF; intramyofibrillar, Intra; and subsarcolemmal, SS) during repeated high‐intensity intermittent exercise. Eighteen moderately to well‐trained male participants performed three periods of 10 × 45 s cycling at ∼105% watt max (EX1–EX3) coupled with 5 × 6 s maximal sprints at baseline and after each period. Muscle biopsies were sampled at baseline and after EX1 and EX3. A higher glycogen breakdown rate in type 2 compared to type 1 fibres was found during EX1 for the Intra (−72 vs. −45%) and IMF (−59 vs. −35%) glycogen fractions (P < 0.001) but with no differences for SS glycogen (−52 vs. −40%). In contrast, no fibre type differences were observed during EX2–EX3, where the utilisation of Intra and IMF glycogen in type 2 fibres was reduced, resulting in depletion of all three subcellular fractions to very low levels post‐exercise within both fibre types. Importantly, large heterogeneity in single‐fibre glycogen utilisation was present with an early depletion of especially Intra glycogen in individual type 2 fibres. In conclusion, there is a clear fibre type‐ and localisation‐specific glycogen utilisation during high‐intensity intermittent exercise, which varies with time course of exercise and is characterised by exacerbated pool‐specific glycogen depletion at the single‐fibre level. image Key points: Muscle glycogen is the major fuel during high‐intensity exercise and is stored in distinct subcellular areas of the muscle cell in close vicinity to the main energy consumption sites. In the present study quantitative electron microscopy imaging was used to investigate the utilisation pattern of three distinct subcellular muscle glycogen fractions during repeated high‐intensity intermittent exercise. It is shown that the utilisation differs dependent on fibre type, subcellular localisation and time course of exercise and with large single‐fibre heterogeneity. These findings expand on our understanding of subcellular muscle glycogen metabolism during exercise and may help us explain how reductions in muscle glycogen can attenuate muscle function even at only moderately lowered whole‐muscle glycogen concentrations.
- 18 September 2022
- Published by Wiley in The Journal of Physiology
- Vol. 600 (21) , 4713-4730
Open AccessFibre type‐ and localisation‐specific muscle glycogen utilisation during repeated high‐intensity intermittent exercise
Glycogen particles are situated in key areas of the muscle cell in the vicinity of the main energy‐consumption sites and may be utilised heterogeneously dependent on the nature of the metabolic demands. The present study aimed to investigate the time course of fibre type‐specific utilisation of muscle glycogen in three distinct subcellular fractions (intermyofibrillar, IMF; intramyofibrillar, Intra; and subsarcolemmal, SS) during repeated high‐intensity intermittent exercise. Eighteen moderately to well‐trained male participants performed three periods of 10 × 45 s cycling at ∼105% watt max (EX1–EX3) coupled with 5 × 6 s maximal sprints at baseline and after each period. Muscle biopsies were sampled at baseline and after EX1 and EX3. A higher glycogen breakdown rate in type 2 compared to type 1 fibres was found during EX1 for the Intra (−72 vs. −45%) and IMF (−59 vs. −35%) glycogen fractions (P < 0.001) but with no differences for SS glycogen (−52 vs. −40%). In contrast, no fibre type differences were observed during EX2–EX3, where the utilisation of Intra and IMF glycogen in type 2 fibres was reduced, resulting in depletion of all three subcellular fractions to very low levels post‐exercise within both fibre types. Importantly, large heterogeneity in single‐fibre glycogen utilisation was present with an early depletion of especially Intra glycogen in individual type 2 fibres. In conclusion, there is a clear fibre type‐ and localisation‐specific glycogen utilisation during high‐intensity intermittent exercise, which varies with time course of exercise and is characterised by exacerbated pool‐specific glycogen depletion at the single‐fibre level. image Key points: Muscle glycogen is the major fuel during high‐intensity exercise and is stored in distinct subcellular areas of the muscle cell in close vicinity to the main energy consumption sites. In the present study quantitative electron microscopy imaging was used to investigate the utilisation pattern of three distinct subcellular muscle glycogen fractions during repeated high‐intensity intermittent exercise. It is shown that the utilisation differs dependent on fibre type, subcellular localisation and time course of exercise and with large single‐fibre heterogeneity. These findings expand on our understanding of subcellular muscle glycogen metabolism during exercise and may help us explain how reductions in muscle glycogen can attenuate muscle function even at only moderately lowered whole‐muscle glycogen concentrations.
- Open Access
- 17 February 2022
- Published by Public Library of Science (PLoS) in PLOS Digital Health
- Vol. 1 (2) , e0000014
Transfer learning for non-image data in clinical research: A scoping review
Background: Transfer learning is a form of machine learning where a pre-trained model trained on a specific task is reused as a starting point and tailored to another task in a different dataset. While transfer learning has garnered considerable attention in medical image analysis, its use for clinical non-image data is not well studied. Therefore, the objective of this scoping review was to explore the use of transfer learning for non-image data in the clinical literature. Methods and findings: We systematically searched medical databases (PubMed, EMBASE, CINAHL) for peer-reviewed clinical studies that used transfer learning on human non-image data. We included 83 studies in the review. More than half of the studies (63%) were published within 12 months of the search. Transfer learning was most often applied to time series data (61%), followed by tabular data (18%), audio (12%) and text (8%). Thirty-three (40%) studies applied an image-based model to non-image data after transforming data into images (e.g. spectrograms). Twenty-nine (35%) studies did not have any authors with a health-related affiliation. Many studies used publicly available datasets (66%) and models (49%), but fewer shared their code (27%). Conclusions: In this scoping review, we have described current trends in the use of transfer learning for non-image data in the clinical literature. We found that the use of transfer learning has grown rapidly within the last few years. We have identified studies and demonstrated the potential of transfer learning in clinical research in a wide range of medical specialties. More interdisciplinary collaborations and the wider adaption of reproducible research principles are needed to increase the impact of transfer learning in clinical research.
- 17 February 2022
- Published by Public Library of Science (PLoS) in PLOS Digital Health
- Vol. 1 (2) , e0000014
Open AccessTransfer learning for non-image data in clinical research: A scoping review
Background: Transfer learning is a form of machine learning where a pre-trained model trained on a specific task is reused as a starting point and tailored to another task in a different dataset. While transfer learning has garnered considerable attention in medical image analysis, its use for clinical non-image data is not well studied. Therefore, the objective of this scoping review was to explore the use of transfer learning for non-image data in the clinical literature. Methods and findings: We systematically searched medical databases (PubMed, EMBASE, CINAHL) for peer-reviewed clinical studies that used transfer learning on human non-image data. We included 83 studies in the review. More than half of the studies (63%) were published within 12 months of the search. Transfer learning was most often applied to time series data (61%), followed by tabular data (18%), audio (12%) and text (8%). Thirty-three (40%) studies applied an image-based model to non-image data after transforming data into images (e.g. spectrograms). Twenty-nine (35%) studies did not have any authors with a health-related affiliation. Many studies used publicly available datasets (66%) and models (49%), but fewer shared their code (27%). Conclusions: In this scoping review, we have described current trends in the use of transfer learning for non-image data in the clinical literature. We found that the use of transfer learning has grown rapidly within the last few years. We have identified studies and demonstrated the potential of transfer learning in clinical research in a wide range of medical specialties. More interdisciplinary collaborations and the wider adaption of reproducible research principles are needed to increase the impact of transfer learning in clinical research.
- Open Access
Transfer learning for non-image data in clinical research: a scoping review
Background Transfer learning is a form of machine learning where a pre-trained model trained on a specific task is reused as a starting point and tailored to another task in a different dataset. While transfer learning has garnered considerable attention in medical image analysis, its use for clinical non-image data is not well studied. Therefore, the objective of this scoping review was to explore the use of transfer learning for non-image data in the clinical literature. Methods and Findings We systematically searched medical databases (PubMed, EMBASE, CINAHL) for peer-reviewed clinical studies that used transfer learning on human non-image data. We included 83 studies in the review. More than half of the studies (63%) were published within 12 months of the search. Transfer learning was most often applied to time series data (61%), followed by tabular data (18%), audio (12%) and text (8%). Thirty-three (40%) studies applied an image-based model to non-image data after transforming data into images (e.g. spectrograms). Twenty-nine (35%) studies did not have any authors with a health-related affiliation. Many studies used publicly available datasets (66%) and models (49%), but fewer shared their code (27%). Conclusions In this scoping review, we have described current trends in the use of transfer learning for non-image data in the clinical literature. We found that the use of transfer learning has grown rapidly within the last few years. We have identified studies and demonstrated the potential of transfer learning in clinical research in a wide range of medical specialties. More interdisciplinary collaborations and the wider adaption of reproducible research principles are needed to increase the impact of transfer learning in clinical research.
Open AccessTransfer learning for non-image data in clinical research: a scoping review
Background Transfer learning is a form of machine learning where a pre-trained model trained on a specific task is reused as a starting point and tailored to another task in a different dataset. While transfer learning has garnered considerable attention in medical image analysis, its use for clinical non-image data is not well studied. Therefore, the objective of this scoping review was to explore the use of transfer learning for non-image data in the clinical literature. Methods and Findings We systematically searched medical databases (PubMed, EMBASE, CINAHL) for peer-reviewed clinical studies that used transfer learning on human non-image data. We included 83 studies in the review. More than half of the studies (63%) were published within 12 months of the search. Transfer learning was most often applied to time series data (61%), followed by tabular data (18%), audio (12%) and text (8%). Thirty-three (40%) studies applied an image-based model to non-image data after transforming data into images (e.g. spectrograms). Twenty-nine (35%) studies did not have any authors with a health-related affiliation. Many studies used publicly available datasets (66%) and models (49%), but fewer shared their code (27%). Conclusions In this scoping review, we have described current trends in the use of transfer learning for non-image data in the clinical literature. We found that the use of transfer learning has grown rapidly within the last few years. We have identified studies and demonstrated the potential of transfer learning in clinical research in a wide range of medical specialties. More interdisciplinary collaborations and the wider adaption of reproducible research principles are needed to increase the impact of transfer learning in clinical research.