Jose María Salinas
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
Jose María Salinas has a total of 55 co-authors. Here are the top co-authors by number of shared publications:
- Antonio Pertusa
University Institute for Computing Research, University of Alicante, Alicante, Spain
4 shared publications - Aurelia Bustos
Dpto. Lenguajes y Sistemas Informáticos Universidad de Alicante San Vicente del Raspeig E-03690 Alicante Spain
4 shared publications - María de la Iglesia-Vayá
Unidad Mixta de Imagen Biomédica e Inteligencia Artificial FISABIO-CIPF, Fundación para el Fomento de la Investigación Sanitario y Biomédica de la Comunidad Valenciana, Valencia, Spain
3 shared publications - Emilio Flores
Clinical Laboratory Department, Hospital Universitario de San Juan, 03550 Alicante, Spain
2 shared publications - Germán González
Robotics, Vision and Intelligent Technologies, Department of Computational Sciences and Artificial Intelligence, University of Alicante, Alicante, Spain
2 shared publications - Jara Llenas-García
HIV Unit, Internal Medicine Department, Hospital Universitario La Paz, IdiPAZ , Madrid
2 shared publications
Latest Publications
- Open Access
- 29 October 2025
- Published by MDPI AG in Biomedicines
- Vol. 13 (11) , 2648
Reclassification of Low or Intermediate Cardiovascular Risk by Determining Lipoprotein(a) Levels
Background: Lipoprotein(a) [Lp(a)] is a modifier of cardiovascular risk, and it should be determined at least once in a lifetime. Methods: Subjects with low or moderate cardiovascular risk, estimated by SCORE2, were invited to have a determination of Lp(a), and those with Lp(a) > 50 mg/dL were classified into a higher-risk category. Eligibility of statins was assessed according to treatment targets. Results: We analyzed 140 subjects, with a mean age of 54.3 (8.1) years and 62.9% women. The median Lp(a) was 15.2 (interquartile range: 6.7–44.5) mg/dL, and 22.1% of the cohort had Lp(a) > 50 mg/dL. No differences were observed in mean age, sex, or lipid profile in subjects with Lp(a) below or above 50 mg/dL; alkaline phosphatase (ALP) was significantly higher in subjects with Lp(a) > 50 mg/dL. After incorporating Lp(a) values into the SCORE-2 assessment, 22.6% of individuals initially of low risk were reclassified as moderate risk, and 77.4% were reclassified from moderate to high risk; moreover, 61.4% (86 subjects) were considered eligible for treatment with statins. Conclusions: Our results highlight that 22.1% of the subjects classified as low or moderate cardiovascular risk by SCORE-2 are reclassified to higher risk, and 61.4% were eligible for statin treatment as a result of Lp(a) testing.
- 29 October 2025
- Published by MDPI AG in Biomedicines
- Vol. 13 (11) , 2648
Open AccessReclassification of Low or Intermediate Cardiovascular Risk by Determining Lipoprotein(a) Levels
Background: Lipoprotein(a) [Lp(a)] is a modifier of cardiovascular risk, and it should be determined at least once in a lifetime. Methods: Subjects with low or moderate cardiovascular risk, estimated by SCORE2, were invited to have a determination of Lp(a), and those with Lp(a) > 50 mg/dL were classified into a higher-risk category. Eligibility of statins was assessed according to treatment targets. Results: We analyzed 140 subjects, with a mean age of 54.3 (8.1) years and 62.9% women. The median Lp(a) was 15.2 (interquartile range: 6.7–44.5) mg/dL, and 22.1% of the cohort had Lp(a) > 50 mg/dL. No differences were observed in mean age, sex, or lipid profile in subjects with Lp(a) below or above 50 mg/dL; alkaline phosphatase (ALP) was significantly higher in subjects with Lp(a) > 50 mg/dL. After incorporating Lp(a) values into the SCORE-2 assessment, 22.6% of individuals initially of low risk were reclassified as moderate risk, and 77.4% were reclassified from moderate to high risk; moreover, 61.4% (86 subjects) were considered eligible for treatment with statins. Conclusions: Our results highlight that 22.1% of the subjects classified as low or moderate cardiovascular risk by SCORE-2 are reclassified to higher risk, and 61.4% were eligible for statin treatment as a result of Lp(a) testing.
- Top Cited Papers
- 26 June 2025
- Published by Massachusetts Medical Society in NEJM AI
- Vol. 2 (7)
PadChest-GR: A Bilingual Chest X-Ray Dataset for Grounded Radiology Report Generation
- 26 June 2025
- Published by Massachusetts Medical Society in NEJM AI
- Vol. 2 (7)
Top Cited PapersPadChest-GR: A Bilingual Chest X-Ray Dataset for Grounded Radiology Report Generation
- Open Access
- 15 April 2025
- Published by Springer Nature in Insights into Imaging
- Vol. 16 (1) , 1-10
Classification and segmentation of hip fractures in x-rays: highlighting fracture regions for interpretable diagnosis
Objective: To develop an artificial intelligence (AI) system capable of classifying and segmenting femoral fractures. To compare its performance against existing state-of-the-art methods. Methods: This Institutional Review Board (IRB)-approved retrospective study did not require informed consent. 10,308 hip x-rays from 2618 patients were retrieved from the hospital PACS. 986 were randomly selected for annotation and randomly split into training, validation, and test sets at the patient level. Two radiologists segmented and classified femoral fractures based on their location (femoral neck, pertrochanteric region, or subtrochanteric region) and grade, using the Evans and Garden scales for neck and pertrochanteric regions, respectively. A YOLOv8 segmentation convolutional neural network (CNN) was trained to generate fracture masks and indicate their class and grade. Classification CNNs were trained in the same dataset for method comparison. Results: On the test set, YOLOv8 achieved a Dice coefficient of 0.77 (95% CI: 0.56–0.98) for segmenting fractures, an accuracy of 86.2% (95% CI: 80.77–90.55) for classification and grading, and an AUC of 0.981 (95% CI: 0.965–0.997) for fracture detection. These metrics are on par with or exceed those of previously published AI methods, demonstrating the efficacy of our approach. Conclusions: The high accuracy and AUC values demonstrate the potential of the proposed neural network as a reliable tool in clinical settings. Further, it is the first to provide a precise segmentation of femoral fractures, as indicated by the Dice scores, which may enhance interpretability. A formal evaluation is planned to further assess its clinical applicability. Critical relevance statement: The proposed system offers high granularity in fracture classification and is the first to segment femoral fractures, ensuring interpretability. Key Points: We present the first AI method that segments and grades femoral fractures. The method classifies fractures with fracture location and type. High accuracy and interpretability promise utility in clinical practice. Graphical
- 15 April 2025
- Published by Springer Nature in Insights into Imaging
- Vol. 16 (1) , 1-10
Open AccessClassification and segmentation of hip fractures in x-rays: highlighting fracture regions for interpretable diagnosis
Objective: To develop an artificial intelligence (AI) system capable of classifying and segmenting femoral fractures. To compare its performance against existing state-of-the-art methods. Methods: This Institutional Review Board (IRB)-approved retrospective study did not require informed consent. 10,308 hip x-rays from 2618 patients were retrieved from the hospital PACS. 986 were randomly selected for annotation and randomly split into training, validation, and test sets at the patient level. Two radiologists segmented and classified femoral fractures based on their location (femoral neck, pertrochanteric region, or subtrochanteric region) and grade, using the Evans and Garden scales for neck and pertrochanteric regions, respectively. A YOLOv8 segmentation convolutional neural network (CNN) was trained to generate fracture masks and indicate their class and grade. Classification CNNs were trained in the same dataset for method comparison. Results: On the test set, YOLOv8 achieved a Dice coefficient of 0.77 (95% CI: 0.56–0.98) for segmenting fractures, an accuracy of 86.2% (95% CI: 80.77–90.55) for classification and grading, and an AUC of 0.981 (95% CI: 0.965–0.997) for fracture detection. These metrics are on par with or exceed those of previously published AI methods, demonstrating the efficacy of our approach. Conclusions: The high accuracy and AUC values demonstrate the potential of the proposed neural network as a reliable tool in clinical settings. Further, it is the first to provide a precise segmentation of femoral fractures, as indicated by the Dice scores, which may enhance interpretability. A formal evaluation is planned to further assess its clinical applicability. Critical relevance statement: The proposed system offers high granularity in fracture classification and is the first to segment femoral fractures, ensuring interpretability. Key Points: We present the first AI method that segments and grades femoral fractures. The method classifies fractures with fracture location and type. High accuracy and interpretability promise utility in clinical practice. Graphical
- Open Access
- 16 January 2024
- Published by Omnia Publisher SL in Journal of Industrial Engineering and Management
- Vol. 17 (1) , 168-181
Hospitalization forecast to inform COVID-19 pandemic planning and resource allocation using discrete event simulation
Purpose: This study aims to address the pressing need for accurate forecasting of healthcare resource demands during the COVID-19 pandemic. It presents a novel approach that combines a stochastic Markov model and a discrete event simulation model to dynamically predict hospital admissions and daily occupancy of hospital and ICU beds. Design/methodology/approach: The research builds upon existing work related to predicting COVID-19 spread and patient influx to hospital emergency departments. The proposed model was developed and validated at San Juan de Alicante University Hospital from July 10, 2020, to January 10, 2022, and externally validated at Hospital Vega Baja. The model involves an admissions generator based on a stochastic Markov model, feeding data into a discrete event simulation model in the R programming language. The probabilities of hospital admission were calculated based on age-stratified positive SARS-COV-2 results from the health department's catchment population. The discrete event simulation model simulates distinct patient pathways within the hospital to estimate bed occupancy for the upcoming week. The performance of the model was measured using the median absolute difference (MAD) between predicted and actual demand. Findings: When applied to data from San Juan hospital, the admissions generator demonstrated a MAD of 6 admissions/week (interquartile range [IQR] 2-11). The MAD between the model's predictions and actual bed occupancy was 20 beds/day (IQR 5-43), equivalent to 5% of total hospital beds. For ICU occupancy, the MAD was 4 beds/day (IQR 2-7), constituting 25% of ICU beds. Evaluation with data from Hospital Vega Baja showcased an admissions generator MAD of 2.42 admissions/week (IQR 1.02-7.41). The MAD between the model's predictions and actual bed occupancy was 18 beds/day (IQR 19.57-38.89), approximately 5.1% of hospital beds. The ICU occupancy MAD was 3 beds/day (IQR 1-5), making up 21.4% of ICU beds. Practical implications: The dynamic predictions of hospital admissions, ward beds, and ICU occupancy for COVID-19 patients proved highly valuable to hospital managers, facilitating early and informed planning of resource allocation. Additionally, this study underscores the importance of utilizing simulation techniques to predict and manage hospital occupancy levels, thereby enhancing decision-making in hospital bed management, not only during pandemics but also during regular periods. Originality/value: This study introduces a novel hybrid approach that combines stochastic modeling and discrete event simulation to forecast healthcare resource demands during the COVID-19 pandemic. The methodology's effectiveness in predicting admissions and bed occupancy contributes to improved resource planning and situational awareness.
- 16 January 2024
- Published by Omnia Publisher SL in Journal of Industrial Engineering and Management
- Vol. 17 (1) , 168-181
Open AccessHospitalization forecast to inform COVID-19 pandemic planning and resource allocation using discrete event simulation
Purpose: This study aims to address the pressing need for accurate forecasting of healthcare resource demands during the COVID-19 pandemic. It presents a novel approach that combines a stochastic Markov model and a discrete event simulation model to dynamically predict hospital admissions and daily occupancy of hospital and ICU beds. Design/methodology/approach: The research builds upon existing work related to predicting COVID-19 spread and patient influx to hospital emergency departments. The proposed model was developed and validated at San Juan de Alicante University Hospital from July 10, 2020, to January 10, 2022, and externally validated at Hospital Vega Baja. The model involves an admissions generator based on a stochastic Markov model, feeding data into a discrete event simulation model in the R programming language. The probabilities of hospital admission were calculated based on age-stratified positive SARS-COV-2 results from the health department's catchment population. The discrete event simulation model simulates distinct patient pathways within the hospital to estimate bed occupancy for the upcoming week. The performance of the model was measured using the median absolute difference (MAD) between predicted and actual demand. Findings: When applied to data from San Juan hospital, the admissions generator demonstrated a MAD of 6 admissions/week (interquartile range [IQR] 2-11). The MAD between the model's predictions and actual bed occupancy was 20 beds/day (IQR 5-43), equivalent to 5% of total hospital beds. For ICU occupancy, the MAD was 4 beds/day (IQR 2-7), constituting 25% of ICU beds. Evaluation with data from Hospital Vega Baja showcased an admissions generator MAD of 2.42 admissions/week (IQR 1.02-7.41). The MAD between the model's predictions and actual bed occupancy was 18 beds/day (IQR 19.57-38.89), approximately 5.1% of hospital beds. The ICU occupancy MAD was 3 beds/day (IQR 1-5), making up 21.4% of ICU beds. Practical implications: The dynamic predictions of hospital admissions, ward beds, and ICU occupancy for COVID-19 patients proved highly valuable to hospital managers, facilitating early and informed planning of resource allocation. Additionally, this study underscores the importance of utilizing simulation techniques to predict and manage hospital occupancy levels, thereby enhancing decision-making in hospital bed management, not only during pandemics but also during regular periods. Originality/value: This study introduces a novel hybrid approach that combines stochastic modeling and discrete event simulation to forecast healthcare resource demands during the COVID-19 pandemic. The methodology's effectiveness in predicting admissions and bed occupancy contributes to improved resource planning and situational awareness.
- Open Access
- 1 January 2023
- Published by American Association for the Advancement of Science (AAAS) in Computational and Structural Biotechnology Journal
- Vol. 22, 27-31
Clinical Decision Support systems: A step forward in establishing the clinical laboratory as a decision maker hub - A CDS system protocol implementation in the clinical laboratory
Background: New tools for health information technology have been developed in recent times, such as Clinical Decision Support (CDS) systems, which are any digital solutions designed to help healthcare professionals when making clinical decisions. The study aimed to show how we have adopted a CDS system in the San Juan de Alicante Clinical Laboratory and facilitate the implementation of our protocol in other clinical laboratories. We have user experience and the motivation to improve healthcare tools. The improvement, measurement, and monitoring of interventions and laboratory tests has been our motto for years. Materials and methods: A descriptive research was conducted. All stages in the design of the project are as follows: 1. Set up a multidisciplinary workgroup. 2. Review patients’ data. 3. Identify relevant data from main sources. 4. Design the likely outcomes. 5. Define a complete integration scenario. 6. Monitor and track the impact. To set up this protocol, two new software systems were implemented in our laboratory: AlinIQ CDS v8.2 as Rule Engine, and AlinIQ AIP Integrated Platform v1.6 as Business Intelligence (BI) tool. Results: Our protocol shows the workflow and actions that can be done with a CDS system and also how it could be integrated with other monitoring systems, as well as some examples of KPIs and their outcomes. Conclusions: CDS could be a great strategic asset for clinical laboratories to improve the integration of care, optimize the use of laboratory tests, and add more clinical value to physicians in the interpretation of results.
- 1 January 2023
- Published by American Association for the Advancement of Science (AAAS) in Computational and Structural Biotechnology Journal
- Vol. 22, 27-31
Open AccessClinical Decision Support systems: A step forward in establishing the clinical laboratory as a decision maker hub - A CDS system protocol implementation in the clinical laboratory
Background: New tools for health information technology have been developed in recent times, such as Clinical Decision Support (CDS) systems, which are any digital solutions designed to help healthcare professionals when making clinical decisions. The study aimed to show how we have adopted a CDS system in the San Juan de Alicante Clinical Laboratory and facilitate the implementation of our protocol in other clinical laboratories. We have user experience and the motivation to improve healthcare tools. The improvement, measurement, and monitoring of interventions and laboratory tests has been our motto for years. Materials and methods: A descriptive research was conducted. All stages in the design of the project are as follows: 1. Set up a multidisciplinary workgroup. 2. Review patients’ data. 3. Identify relevant data from main sources. 4. Design the likely outcomes. 5. Define a complete integration scenario. 6. Monitor and track the impact. To set up this protocol, two new software systems were implemented in our laboratory: AlinIQ CDS v8.2 as Rule Engine, and AlinIQ AIP Integrated Platform v1.6 as Business Intelligence (BI) tool. Results: Our protocol shows the workflow and actions that can be done with a CDS system and also how it could be integrated with other monitoring systems, as well as some examples of KPIs and their outcomes. Conclusions: CDS could be a great strategic asset for clinical laboratories to improve the integration of care, optimize the use of laboratory tests, and add more clinical value to physicians in the interpretation of results.
- Open Access
Hospitalization forecast to inform COVID-19 pandemic planning and resource allocation using mathematical models
Background The COVID-19 pandemic has put tremendous pressure on hospital resources around the world. Forecasting demand for healthcare services is important generally, but crucial in epidemic contexts, both to facilitate resource planning and to inform situational awareness. There is abundant research on methods for predicting the spread of COVID-19 and even the arrival of COVID-19 patients to hospitals emergency departments. This study builds on that work to propose a hybrid tool, combining a stochastic Markov model and a discrete event simulation model to dynamically predict hospital admissions and total daily occupancy of hospital and ICU beds. Methods The model was developed and validated at San Juan de Alicante University Hospital from 10 July 2020 to 10 January 2022 and externally validated at Hospital Vega Baja. An admissions generator was developed using a stochastic Markov model that feeds a discrete event simulation model in R. Positive microbiological SARS-COV-2 results from the health department’s catchment population were stratified by patient age to calculate the probabilities of hospital admission. Admitted patients follow distinct pathways through the hospital, which are simulated by the discrete event simulation model, allowing administrators to estimate the bed occupancy for the next week. The median absolute difference (MAD) between predicted and actual demand was used as a model performance measure. Results With respect to the San Juan hospital data, the admissions generator yielded a MAD of 6 admissions/week (interquartile range [IQR] 2-11). The MAD between the tool’s predictions and actual bed occupancy was 20 beds/day (IQR 5-43), or 5% of the hospital beds. The MAD between the intensive care unit (ICU)’s predicted and actual occupancy was 4 beds/day (IQR 2-7), or 25% of the beds. When the model was further evaluated with data from Hospital Vega Baja, the admissions generator showed a MAD of 2.42 admissions/week (IQR 1.02-7.41). The MAD between the tools’ predictions and the actual bed occupancy was 18 beds/day (IQR 19.57-38.89), or 5.1% of the hospital beds. For ICU beds, the MAD was 3 beds/day (IQR 1-5), or 21.4% of the ICU beds. Conclusion Predictions of hospital admissions, ward beds, and ICU occupancy for COVID-19 patients were very useful to hospital managers, allowing early planning of hospital resource allocation.
Open AccessHospitalization forecast to inform COVID-19 pandemic planning and resource allocation using mathematical models
Background The COVID-19 pandemic has put tremendous pressure on hospital resources around the world. Forecasting demand for healthcare services is important generally, but crucial in epidemic contexts, both to facilitate resource planning and to inform situational awareness. There is abundant research on methods for predicting the spread of COVID-19 and even the arrival of COVID-19 patients to hospitals emergency departments. This study builds on that work to propose a hybrid tool, combining a stochastic Markov model and a discrete event simulation model to dynamically predict hospital admissions and total daily occupancy of hospital and ICU beds. Methods The model was developed and validated at San Juan de Alicante University Hospital from 10 July 2020 to 10 January 2022 and externally validated at Hospital Vega Baja. An admissions generator was developed using a stochastic Markov model that feeds a discrete event simulation model in R. Positive microbiological SARS-COV-2 results from the health department’s catchment population were stratified by patient age to calculate the probabilities of hospital admission. Admitted patients follow distinct pathways through the hospital, which are simulated by the discrete event simulation model, allowing administrators to estimate the bed occupancy for the next week. The median absolute difference (MAD) between predicted and actual demand was used as a model performance measure. Results With respect to the San Juan hospital data, the admissions generator yielded a MAD of 6 admissions/week (interquartile range [IQR] 2-11). The MAD between the tool’s predictions and actual bed occupancy was 20 beds/day (IQR 5-43), or 5% of the hospital beds. The MAD between the intensive care unit (ICU)’s predicted and actual occupancy was 4 beds/day (IQR 2-7), or 25% of the beds. When the model was further evaluated with data from Hospital Vega Baja, the admissions generator showed a MAD of 2.42 admissions/week (IQR 1.02-7.41). The MAD between the tools’ predictions and the actual bed occupancy was 18 beds/day (IQR 19.57-38.89), or 5.1% of the hospital beds. For ICU beds, the MAD was 3 beds/day (IQR 1-5), or 21.4% of the ICU beds. Conclusion Predictions of hospital admissions, ward beds, and ICU occupancy for COVID-19 patients were very useful to hospital managers, allowing early planning of hospital resource allocation.
- Top Cited Papers
- 1 December 2020
- Published by Elsevier in Medical Image Analysis
- Vol. 66, 101797
PadChest: A large chest x-ray image dataset with multi-label annotated reports
- 1 December 2020
- Published by Elsevier in Medical Image Analysis
- Vol. 66, 101797
Top Cited PapersPadChest: A large chest x-ray image dataset with multi-label annotated reports
- Open Access
BIMCV COVID-19+: a large annotated dataset of RX and CT images from COVID-19 patients
This paper describes BIMCV COVID-19+, a large dataset from the Valencian Region Medical ImageBank (BIMCV) containing chest X-ray images CXR (CR, DX) and computed tomography (CT) imaging of COVID-19+ patients along with their radiological findings and locations, pathologies, radiological reports (in Spanish), DICOM metadata, Polymerase chain reaction (PCR), Immunoglobulin G (IgG) and Immunoglobulin M (IgM) diagnostic antibody tests. The findings have been mapped onto standard Unified Medical Language System (UMLS) terminology and cover a wide spectrum of thoracic entities, unlike the considerably more reduced number of entities annotated in previous datasets. Images are stored in high resolution and entities are localized with anatomical labels and stored in a Medical Imaging Data Structure (MIDS) format. In addition, 10 images were annotated by a team of radiologists to include semantic segmentation of radiological findings. This first iteration of the database includes 1,380 CX, 885 DX and 163 CT studies from 1,311 COVID-19+ patients. This is, to the best of our knowledge, the largest COVID-19+ dataset of images available in an open format. The dataset can be downloaded from http://bimcv.cipf.es/bimcv-projects/bimcv-covid19.
Open AccessBIMCV COVID-19+: a large annotated dataset of RX and CT images from COVID-19 patients
This paper describes BIMCV COVID-19+, a large dataset from the Valencian Region Medical ImageBank (BIMCV) containing chest X-ray images CXR (CR, DX) and computed tomography (CT) imaging of COVID-19+ patients along with their radiological findings and locations, pathologies, radiological reports (in Spanish), DICOM metadata, Polymerase chain reaction (PCR), Immunoglobulin G (IgG) and Immunoglobulin M (IgM) diagnostic antibody tests. The findings have been mapped onto standard Unified Medical Language System (UMLS) terminology and cover a wide spectrum of thoracic entities, unlike the considerably more reduced number of entities annotated in previous datasets. Images are stored in high resolution and entities are localized with anatomical labels and stored in a Medical Imaging Data Structure (MIDS) format. In addition, 10 images were annotated by a team of radiologists to include semantic segmentation of radiological findings. This first iteration of the database includes 1,380 CX, 885 DX and 163 CT studies from 1,311 COVID-19+ patients. This is, to the best of our knowledge, the largest COVID-19+ dataset of images available in an open format. The dataset can be downloaded from http://bimcv.cipf.es/bimcv-projects/bimcv-covid19.
- Open Access
PadChest: A large chest x-ray image dataset with multi-label annotated reports
We present a labeled large-scale, high resolution chest x-ray dataset for the automated exploration of medical images along with their associated reports. This dataset includes more than 160,000 images obtained from 67,000 patients that were interpreted and reported by radiologists at Hospital San Juan Hospital (Spain) from 2009 to 2017, covering six different position views and additional information on image acquisition and patient demography. The reports were labeled with 174 different radiographic findings, 19 differential diagnoses and 104 anatomic locations organized as a hierarchical taxonomy and mapped onto standard Unified Medical Language System (UMLS) terminology. Of these reports, 27% were manually annotated by trained physicians and the remaining set was labeled using a supervised method based on a recurrent neural network with attention mechanisms. The labels generated were then validated in an independent test set achieving a 0.93 Micro-F1 score. To the best of our knowledge, this is one of the largest public chest x-ray database suitable for training supervised models concerning radiographs, and the first to contain radiographic reports in Spanish. The PadChest dataset can be downloaded from http://bimcv.cipf.es/bimcv-projects/padchest/.
Open AccessPadChest: A large chest x-ray image dataset with multi-label annotated reports
We present a labeled large-scale, high resolution chest x-ray dataset for the automated exploration of medical images along with their associated reports. This dataset includes more than 160,000 images obtained from 67,000 patients that were interpreted and reported by radiologists at Hospital San Juan Hospital (Spain) from 2009 to 2017, covering six different position views and additional information on image acquisition and patient demography. The reports were labeled with 174 different radiographic findings, 19 differential diagnoses and 104 anatomic locations organized as a hierarchical taxonomy and mapped onto standard Unified Medical Language System (UMLS) terminology. Of these reports, 27% were manually annotated by trained physicians and the remaining set was labeled using a supervised method based on a recurrent neural network with attention mechanisms. The labels generated were then validated in an independent test set achieving a 0.93 Micro-F1 score. To the best of our knowledge, this is one of the largest public chest x-ray database suitable for training supervised models concerning radiographs, and the first to contain radiographic reports in Spanish. The PadChest dataset can be downloaded from http://bimcv.cipf.es/bimcv-projects/padchest/.