A Labeled Dataset for Osteoporosis Screening Based on Electromagnetic Attenuation

Dionísio D. A. Carvalho^1*Mikael M. R. Costa da Silva¹Gabriela A. Albuquerque¹Agnaldo S. Cruz¹Felipe Fernandes^1*Ingridy Barbalho¹João Paulo Q. Santos^1,2Antonio Higor Freire de Morais^1,2Karilany Dantas Coutinho¹ANTONIO CAMPOS³Paulo Gil^4,5Cesar Teixeira⁴Jorge Henriques⁴Guilherme M. Machado⁶Ricardo A. M. Valentim¹

• ¹Laboratory of Technological Innovation in Health (LAIS), Federal University of Rio Grande do Norte (UFRN), Natal, Rio Grande do Norte, Brazil, Natal, Rio Grande do Norte, Brazil
• ²Advanced Nucleus of Technological Innovation (NAVI), Federal Institute of Rio Grande do Norte (IFRN), Natal/RN, Brazil, Natal, Rio de Janeiro, Brazil
• ³Electrical and Computer Engineering Graduate Program, Federal University of Rio Grande do Norte, Natal, RN, Brazil
• ⁴University of Coimbra, Center for Informatics and Systems of the University of Coimbra, Department of Informatics Engineering, Coimbra, Portugal
• ⁵CTS-UNINOVA, Department of Electrical and Computer Engineering, NOVA School of Science and Technology, Campus de Caparica, Caparica, Portugal
• ⁶Research Department, ECE-Engineering School, Paris, France

Osteoporosis is an asymptomatic disease characterized by bone microarchitecture deterioration due to multiple risk factors, such as low bone mineral density (BMD), nutritional deficiency, a sedentary lifestyle, alcohol use, smoking, genetic factors, and the use of medications such as glucocorticoids, for example (8,16,1,11). The worldwide prevalence of osteoporosis is estimated to be 23.1% in women and 11.7% in men (17). The economic burden related to osteoporosis fractures is significant, costing approximately $17.9 billion a year in the US alone (7).Studies show that the early and systematic identification of people with clinical indicators of osteoporosis, combined with primary care, appropriate interventions, and the use of therapeutic drugs has reduced the risk of fractures (12,5). Clark and colleagues (6) demonstrated in a randomized clinical trial that primary healthcare screening tools increased osteoporosis drug prescriptions by 124% and reduced the incidence of fractures in the observed group. Their results support the premise that improving preventive screening methods is essential for identifying individuals at high risk of osteoporotic fractures. Therefore, the development of low-cost digital health technologies or solutions for rapid diagnosis in primary health care is essential.Early screening for osteoporosis is a challenging problem because the most appropriate diagnostic method, based on dual-energy X-ray absorptiometry (DXA) to measure the number of grams of mineral per square centimeter (g/cm 2 ) (from the lumbar spine, total femur, femoral neck or middle third of the radius) (13,4), is highly expensive. Furthermore, as this method uses ionizing radiation technology and requires large, specialized equipment, these devices are generally concentrated in major healthcare centers. Given Brazil's social inequality and geographical challenges, access to such services is often bureaucratic, slow, and, in some cases, unfeasible.To overcome these challenges Brazilian researchers have developed a low-cost, portable device for osteoporosis screening called OSSEUS (10,9). The device combines techniques for measuring the attenuation of electromagnetic waves passing through bone tissue, extracting patient features(risk factors), and recognizing patterns to help classify osteoporosis and provide decision support for healthcare professionals (15). A primary health care screening study using OSSEUS with a group of 505 people referred for a DXA scan found that 78.2% could start preventive care. In addition, the study also showed that 110 people (21.8%) were healthy (concerning this pathology) and did not need to be referred to specialized health centers to get a DXA screening (2).Recognizing the importance of osteoporosis screening in primary health care and early care, such as treatment to reduce the incidence of fractures -especially in people identified by OSSEUS as having risk factors for the disease -, this study aims to provide a database of 669 people available to support future research, especially in the field of artificial intelligence and machine learning, and contribute to the development of digital health solutions in response to osteoporosis. The dataset is available at: https://doi.org/10.5281/zenodo.14259374.This study consists of a data report conducted and guided by a descriptive analysis of osteoporosis and innovative technologies enabling early disease detection. The dataset comprises 669 people (575 females, 94 males), 156 from the control group, aged between 20-85 years (median age: 55; mean age: 54.5 ± 13.2), and 513 with low BMD, aged between 18-101 years (median age: 66; mean age: 65.3 ± 11.2). Table 1 shows the demographic, anthropometric, risk factor, OSSEUS, and DXA features of the population in the dataset. All participants in the dataset were volunteers who met the eligibility criteria. Participation was limited to individuals of both sexes, aged 18 or older, with a medical indication (prescription/request) for DXA. In addition to these criteria, participants were required to have intact middle finger phalanges to 58 meet specific prerequisites for OSSEUS. The data from the 669 people was collected at the Onofre Lopes University Hospital (HUOL) at the The data were inspected for instances with missing values or values outside the normal range (outliers).Instances with missing values for attributes related to age and the DXA scans were removed. Next, the data generated by the OSSEUS device was inspected and the instances with attenuation values exceeding those obtained during calibration were removed -since results falling outside the calibrated range may indicate measurement errors or malfunctioning of the device, compromising the validity and quality of the readings.A new attribute, named' worst deviation', was defined to store the lowest value among the four possible sites DXA can be performed. This attribute, along with age, gender, and menopause attributes, is considered in calculations for a diagnostic definition of metabolic bone diseases, as recommended by the Brazilian Ministry of Health in its Clinical Protocol and Therapeutic Guidelines for Osteoporosis (PCDT) (4). In summary, based on the calculation of clinical variables, menopausal women and individuals aged 50 or older are diagnosed using the T-score deviation. The remaining registries use the Z-score for diagnosis (3).Deviation values below -2.0 in the Z-score or below -1.0 in the T-score are classified as "low BMD" or "osteoporosis" (14).The dataset also enables feature engineering, i.e. the definition, creation, or aggregation of predictive attributes. For instance, it is possible to calculate the Body Mass Index (BMI) from Electronic Health Record (EHR) data, considering body weight (kg), height (m), and the following equation: BM I = weight height 2 .In addition, it is possible to conduct investigative research using OSSEUS data, considering the real or percentage differences between the calibration and attenuation attributes of the biomedical device and the influence of obstacles in this context. The obstacle can be determined by estimating the area or volume of the medial phalanx of the middle finger, using the following equations: area = 2πr(r + h); volume = πr 2 h, where:d = medial height+medial width, determines the diameter of the medial phalanx of the middle finger;r = d 2 , defines the medial phalanx radius of the middle finger;h = medial length, represents the length of the medial phalanx of the middle finger.The dataset presents relevant characteristics of a particular region of Brazil, which can significantly