Retrospective evaluation and dating of non-accidental rib fractures in infants

High sensitivity methods for automated rib fracture detection in pediatric radiographs

Rib fractures are highly predictive of non-accidental trauma in children under 3 years old. Rib fracture detection in pediatric radiographs is challenging because fractures can be obliquely oriented to the imaging detector, obfuscated by other structures, incomplete, and non-displaced. Prior studies have shown up to two-thirds of rib fractures may be missed during initial interpretation. In this paper, we implemented methods for improving the sensitivity (i.e. recall) performance for detecting and localizing rib fractures in pediatric chest radiographs to help augment performance of radiology interpretation. These methods adapted two convolutional neural network (CNN) architectures, RetinaNet and YOLOv5, and our previously proposed decision scheme, "avalanche decision", that dynamically reduces the acceptance threshold for proposed regions in each image. Additionally, we present contributions of using multiple image pre-processing and model ensembling techniques. Using a custom dataset of 1109 pediatric chest radiographs manually labeled by seven pediatric radiologists, we performed 10-fold cross-validation and reported detection performance using several metrics, including F2 score which summarizes precision and recall for high-sensitivity tasks. Our best performing model used three ensembled YOLOv5 models with varied input processing and an avalanche decision scheme, achieving an F2 score of 0.725 ± 0.012. Expert inter-reader performance yielded an F2 score of 0.732. Results demonstrate that our combination of sensitivity-driving methods provides object detector performance approaching the capabilities of expert human readers, suggesting that these methods may provide a viable approach to identify all rib fractures.

A Patch-Based Deep Learning Approach for Detecting Rib Fractures on Frontal Radiographs in Young Children

Chest radiography is the modality of choice for the identification of rib fractures in young children and there is value for the development of computer-aided rib fracture detection in this age group. However, the automated identification of rib fractures on chest radiographs can be challenging due to the need for high spatial resolution in deep learning frameworks. A patch-based deep learning algorithm was developed to automatically detect rib fractures on frontal chest radiographs in children under 2 years old. A total of 845 chest radiographs of children 0-2 years old (median: 4 months old) were manually segmented for rib fractures by radiologists and served as the ground-truth labels. Image analysis utilized a patch-based sliding-window technique, to meet the high-resolution requirements for fracture detection. Standard transfer learning techniques used ResNet-50 and ResNet-18 architectures. Area-under-curve for precision-recall (AUC-PR) and receiver-operating-characteristic (AUC-ROC), along with patch and whole-image classification metrics, were reported. On the test patches, the ResNet-50 model showed AUC-PR and AUC-ROC of 0.25 and 0.77, respectively, and the ResNet-18 showed an AUC-PR of 0.32 and AUC-ROC of 0.76. On the whole-radiograph level, the ResNet-50 had an AUC-ROC of 0.74 with 88% sensitivity and 43% specificity in identifying rib fractures, and the ResNet-18 had an AUC-ROC of 0.75 with 75% sensitivity and 60% specificity in identifying rib fractures. This work demonstrates the utility of patch-based analysis for detection of rib fractures in children under 2 years old. Future work with large cohorts of multi-institutional data will improve the generalizability of these findings to patients with suspicion of child abuse.

Dating birth-related clavicular fractures: pediatric radiologists versus artificial intelligence

BACKGROUND

Fracture dating from skeletal surveys is crucial in the diagnosis and investigation of infant abuse. However, this task is challenging because of the subjective nature of the radiologic interpretation and the lack of ground truth. Researchers have used birth-related clavicle fractures as a surrogate to study the radiographic pattern of healing; however, they did not elucidate the accuracy performance of the radiologists in dating fractures.

OBJECTIVE

To determine the accuracy of radiologists in dating birth-related clavicle fractures and compare their performance to that achieved by computer algorithm.

MATERIALS AND METHODS

We used a previously assembled birth-related clavicle fracture database consisting of 416 anteroposterior clavicle radiographs as the study cohort. The average and standard deviation of the fracture age within this database were 24 days and 18 days, respectively. Three blinded radiologists independently estimated the ages of the clavicle fractures depicted in the radiographs within the database. We compared these estimation results to those made by a recently published deep-learning (DL) model conducted with the identical infant cohort. We calculated standard error metrics to compare the accuracy performances of the radiologists and the computer model.

RESULTS

The intra- and inter-reader agreements of the fracture age estimates by the radiologists were moderate to good. The radiologists estimated the fracture ages with a mean absolute error (MAE) of 6.1-7.1 days, and standard deviation of the absolute error of 6.3-8.3 days. The accuracy performances of the three radiologists were not significantly different from one another. In comparison, the DL model estimated the age of clavicle fractures with an MAE of 4.2 days, significantly lower than all of the radiologists (P < 0.001).

CONCLUSION

Three experienced pediatric radiologists dated clavicular fractures with moderate-good intra- and inter-reader agreements. The correlations between the radiologists' estimates and the ground truth were moderate to good. The fracture ages assigned by the DL model showed superior correlation with the ground truth compared to radiologists' dating estimates.

Thoracic trauma: Clinical and paleopathological perspectives

OBJECTIVES

Although trauma is one of the most significant areas of study in paleopathology, most studies focus on fractures of single anatomical elements. Paleopathological research on regional trauma, such as of the thorax, is rare. This paper explores the causes, complications, and consequences of adult thoracic trauma using clinical data in order to inform paleopathological research.

MATERIALS AND METHODS

Trends in paleopathological thoracic trauma literature were assessed by evaluating publications from Bioarchaeology International, International Journal of Osteoarchaeology, International Journal of Paleopathology, and American Journal of Biological Anthropology. Clinical publications on thoracic trauma throughout time were also assessed through a PubMed search, and modern prevalence data was found through trauma databases such as the National Trauma Databank.

RESULTS

Consideration of thoracic trauma involving concomitant injuries is a recent trend in clinical literature and patient care, but paleopathological research on thoracic trauma has been limited. Since thoracic fractures tend to occur in conjunction with other injuries, assessing them together is critical to the interpretation of trauma in the past.

CONCLUSIONS

Clinical research into thoracic fractures and concomitant injuries provides valuable data for paleopathological research. Evaluating the likelihood and consequences of concomitant injury in skeletal remains provides a more robust understanding of trauma in the past and its impact on past lifeways.

SIGNIFICANCE

This paper provides a review of current clinical and paleopathological literature on thoracic trauma and demonstrates the importance of moving beyond the analysis of fractures or trauma of single anatomical elements.

LIMITATIONS

Thoracic bones are often taphonomically altered and differentially preserved leading to difficulty in identifying and interpreting fractures.

SUGGESTIONS FOR FURTHER RESEARCH

Practical application of the data presented here to archaeological samples will help to advance paleopathological understandings of thoracic trauma.

Deep learning of birth-related infant clavicle fractures: a potential virtual consultant for fracture dating

BACKGROUND

In infant abuse investigations, dating of skeletal injuries from radiographs is desirable to reach a clear timeline of traumatic events. Prior studies have used infant birth-related clavicle fractures as a surrogate to develop a framework for dating of abuse-related fractures.

OBJECTIVE

To develop and train a deep learning algorithm that can accurately date infant birth-related clavicle fractures.

MATERIALS AND METHODS

We modified a deep learning model initially designed for face-age estimation to date infant clavicle fractures. We conducted a computerized search of imaging reports and other medical records at a tertiary children's hospital to identify radiographs of birth-related clavicle fracture in infants ≤ 3 months old (July 2003 to March 2021). We used the resultant database for model training, validation and testing. We evaluated the performance of the deep learning model via a four-fold cross-validation procedure, and calculated accuracy metrics: mean absolute error (MAE), root mean square error (RMSE), intraclass correlation coefficient (ICC) and cumulative score.

RESULTS

The curated database consisted of 416 clavicle radiographs from 213 infants. Average chronological age (equivalent to fracture age) at time of imaging was 24 days. This model estimated the ages of the clavicle fractures with MAE of 4.2 days, RMSE of 6.3 days and ICC of 0.919. On average, 83.7% of the fracture age estimates were accurate to within 7 days of the ground truth.

CONCLUSION

Our deep learning study provides encouraging results for radiographic dating of infant clavicle fractures. With further development and validation, this model might serve as a virtual consultant to radiologists estimating fracture ages in cases of suspected infant abuse.

Postmortem and Antemortem Forensic Assessment of Pediatric Fracture Healing from Radiographs and Machine Learning Classification

Simple Summary

Being able to estimate from X-rays alone how long ago a child’s bone was fractured is important for prosecuting suspected child abuse of living or dead children. This estimate can also help identify a child when all that remains are bones. Experts use various indicators to make these estimates of the age of healing and fully healed fractures, in living and deceased persons, even years after the injury occurred. However, it is not a precise science. We proposed a method using a new combination of indicators to classify fracture healing in children and adolescents. We tested its accuracy with a public database of X-rays of children’s fractures taken during the treatment and healing process. We used part of the X-ray database for training artificial intelligence (AI, or machine learning) programs to classify stages of bone healing when using our new system. We used another portion of the same database to test the performance of the AI system that had been trained with our new classification system. Our new system addresses certain classification ambiguities of a currently used system and is similar in accuracy.

Abstract

A timeline of pediatric bone healing using fracture healing characteristics that can be assessed solely using radiographs would be practical for forensic casework, where the fracture event may precede death by days, months, or years. However, the dating of fractures from radiographs is difficult, imprecise, and lacks consensus, as only a few aspects of the healing process are visible on radiographs. Multiple studies in both the clinical and forensic literature have attempted to develop a usable scale to assess pediatric bone healing on radiographs using various healing characteristics. In contrast to the orthopedic definition, a fracture in forensic casework is only considered to be healed when the area around the fracture has been remodeled to the point that the fracture is difficult to detect on a radiograph or on the surface of the bone itself, a process that can take several years. We subjectively assessed visible characteristics of healing in radiograms of fractures occurring in 942 living children and adolescents. By dividing these assessments into learning and test (validation) sets, the accuracy of a newly proposed fracture healing scale was compared to a previous study. Two machine learning models were used to test predictions of the new scale. All three models produced similar estimates with substantial imprecision. Results corroborate the Malone model with an independent dataset and support the efficacy of using less complex models to estimate fracture age in children.

The Repository of Antemortem Injury Response (REPAIR): an online database for skeletal injuries of known ages

An accurate assessment of time since fracture is an essential component of abuse and death investigations; however, little evidence-based research exists on dating fractures, especially those of the cranial vault. This is primarily due to difficulties in procuring human fracture specimens of known posttraumatic survival times. The aim of this article is to introduce a new database through which limitations imposed by sample procurement may be mitigated. The Repository of Antemortem Injury Response (REPAIR) is a digitally accessible database of cranial vault fractures of known ages with extensive contextual information and visual documentation in the form of photography, radiography, and histological photomicrographs. This repository is a multifunctional tool that serves as a case submission portal for cranial fractures of known posttraumatic survival time, a sample database for research on fracture healing and rates of repair, a resource for comparative assessments of cranial fractures in forensic casework, and an educational tool for healing fracture histomorphology.

Validating scoring systems for fracture healing in infants and young children: pilot study

BACKGROUND

Recent studies have analysed birth-related clavicular fractures to propose time frames for healing that could be applied to dating of all fractures in cases of suspected child abuse.

OBJECTIVE

To assess differences in healing rates between femoral fractures and birth-related clavicular fractures in infants and young children.

MATERIALS AND METHODS

A retrospective 5-year pilot study of femoral fractures in children younger than 3 years of age was performed. Anonymised radiographs were independently scored by two radiologists for stages of fracture healing. In cases of reader disagreement, radiographs were independently scored by a third radiologist.

RESULTS

In total, 74 radiographs (30 children) met the inclusion criteria. Fracture healing evolved over time with subperiosteal new bone formation (SPNBF) appearing first, followed by callus then remodelling. A power calculation for a single proportion, with a level of confidence of 95% and a margin of error of 5%, showed that in a definitive study, 359 radiographs would be required.

CONCLUSION

Although the overall pattern of healing is similar, in this small pilot study, the earliest times for SPNBF and callus formation in femoral fractures appeared to lag behind healing of birth-related clavicular fractures. Remodelling appeared earlier than remodelling of clavicular fractures. A power calculation has determined numbers of femoral radiographs (359) required for a definitive study.

Radiographic timelines for pediatric healing fractures: a systematic review

Skeletal fractures, a common injury in physically abused children, often go undetected and untreated for significant lengths of time and are sometimes incidentally discovered radiographically. Our objective was to review current literature for scientific studies of pediatric fracture healing with associated timelines. We conducted a search of Embase, EBSCOhost, MEDLINE (PubMed), and Web of Science for literature published from the earliest available up to August 2018. We evaluated the included articles for quality, with consideration for use in clinical and forensic settings. Of a total of 313 full-text articles evaluated, 10 met study inclusion criteria. The patient age range among studies was 0-17 years, with children younger than 1 year included in the majority of studies. The fracture locations included in studies were primarily fractures of the upper limb and pectoral girdle, followed by fractures of the lower limb. The radiographic features of healing varied greatly among the studies. Timelines of common fracture healing variables differed significantly among studies. Scientific, radiographic studies of pediatric fracture healing are limited. Gaps in knowledge regarding fracture healing highlight the need for future research and validation studies. Fracture healing timelines derived from existing timelines should be used with caution.

Three-dimensional printed models of the rib cage in children with non-accidental injury as an effective visual-aid tool

Three-dimensional (3-D) printing is gaining terrain in medical education, presurgical evaluation and recently as forensic evidence in court. Physicians, including radiologists, often provide expert testimony in court cases involving children with rib fractures and other injuries concerning for child physical abuse. Effectively communicating the complexities of fractures and other skeletal findings to nonmedical personnel using standard radiology studies can be challenging, especially during medical courtroom testimony. For this reason, we printed two 3-D models of the rib cage from the chest computed tomography (CT) scans of two patients with suspected non-accidental injury. The patients also had available chest radiographs. The DICOM (Digital Imaging and Communications in Medicine) data were 3-D reconstructed and segmented using two attenuation thresholds. We removed unwanted structures and printed them on a commercially available scanner. A pediatric radiologist, blinded to clinical data, reviewed both 3-D models, identified all rib lesions and classified them according to their healing stage. We compared the 3-D models and the chest radiograph against the chest CT as the standard of care. We convened a meeting with the Child Protection Team at out institution to get their feedback and opinions about the models. From our observations of our experts, three spontaneous interactions were observed. Instinctively, the experts picked up and grasped the models, rotating them, feeling them and angling them to better visualize the fractures from multiple angles. The experts expressed a willingness to consider using the models in court.

Subperiosteal new bone formation with the distal tibial classic metaphyseal lesion: prevalence on radiographic skeletal surveys

BACKGROUND

The classic metaphyseal lesion (CML) is a strong indicator of infant abuse, and the distal tibia is one of the most common sites for this injury.

OBJECTIVE

To determine the prevalence of subperiosteal new bone formation accompanying the distal tibial CMLs identified on infant skeletal surveys.

MATERIALS AND METHODS

Skeletal surveys performed for suspected infant abuse (2005-2017) were reviewed. Inclusion criteria were 1) anteroposterior (AP) and lateral radiographs of a distal tibial CML from the initial survey, 2) AP radiograph from the 2-week follow-up survey, 3) additional fractures, 4) child protection team consults and 5) mandated report filing for suspected abuse. We identified 22 distal tibial CMLs from 16 infants. Radiographs of these lesions were shown on the picture archiving and communication system to two blinded pediatric radiologists. Readers indicated the presence/absence of subperiosteal new bone formation on individual and combinations of images.

RESULTS

Inter-reader agreements were fair (kappa=0.47). The prevalence of subperiosteal new bone formation on initial AP radiograph was 34%. Significant increases in the prevalence were found with the addition of follow-up AP radiograph (57%; P<0.001), initial lateral radiograph (57%; P=0.002) and follow-up AP plus initial lateral radiographs (71%; P<0.001). Statistically significant increases in prevalence were also noted when the third view was added to the other two views (increase of 14%; P=0.024).

CONCLUSION

Even when skeletal surveys include initial AP, lateral and follow-up AP radiographs of the tibia, nearly one-third of distal tibial CMLs will fail to demonstrate subperiosteal new bone formation.

Determining when a fracture occurred: Does the method matter? Analysis of the similarity of three different methods for estimating time since fracture of juvenile long bones

Radiographic fracture date estimation is a critical component of skeletal trauma analysis in the living. Several timetables have been proposed for how the appearance of radiographic features can be interpreted to provide a likely time frame for fracture occurrence. This study compares three such timetables for pediatric fractures, by Islam et al. (2000), Malone et al. (2011), and Prosser et al. (2012), in order to determine whether the fracture date ranges produced by using these methods are in agreement with one another. Fracture date ranges were estimated for 112 long bone fractures in 96 children aged 1-17 years, using the three different timetables. The extent of similarity of the intervals was tested by statistically comparing the overlap between the ranges. Results showed that none of the methods were in perfect agreement with one another. Differences seen included the size of the estimated date range for when a fracture occurred, and the specific dates given for both the upper and lower ends of the fracture date range. There was greater similarity between the ranges produced by Malone et al. (2011) and both the other two studies than there was between Islam et al. (2000) and Prosser et al. (2012). The greatest similarity existed between Malone et al. (2011) and Islam et al. (2000). The extent of differences between methods can vary widely, depending on the fracture analysed. Using one timetable gives an average earliest possible fracture date of less than 2 days before another, but the range was extreme, with one method estimating minimum time since fracture as 25 days before another method for a given fracture. In most cases, one method gave maximum time since fracture as a week less than the other two methods, but range was extreme and some estimates were nearly two months different. The variability in fracture date estimates given by these timetables indicates that caution should be exercised when estimating the timing of a juvenile fracture if relying solely on one of the published guides. Future research should be undertaken to compare these methods on a population of known fracture timing, and to better understand the relationship between age of the individual, skeletal health, fracture healing rates, and radiographic characteristics of fracture healing.

Radiological features of healing in newborn clavicular fractures

OBJECTIVES

Given the frequency of abusive fractures among infants, and the lack of research and or evidence for the phases of fracture healing seen in this age group, this study aims to describe a timetable of radiological features of fracture healing among infants in the first months of life.

METHODS

We completed a retrospective cross-sectional time-series study of birth-related clavicle fractures from 2006-2013. A total of 108 digital images were available for review from 61 infants. The presence or absence of four features of healing including periosteal reaction, callus formation, bridging callus and remodelling were scored by three radiologists.

RESULTS

The level of agreement between the radiologists was good to high (0.60-0.90). Features of healing were first seen at 7 days (periosteal reaction), 11 days (callus), 20 days (bridging) and 35 days (remodelling), respectively. The peak periods that each feature was present are as follows: periosteal reaction 11-42 days, callus 12-61 days, bridging 22-63 days and remodelling 49-59 days.

CONCLUSIONS

Birth-associated clavicle fractures in infants follow a logical progression of healing changes. Understanding the expected progression and timing of fracture healing may be helpful as it pertains to the timing of injury in cases of abuse in infants.

KEY POINTS

• Large study describing the time frames of fracture healing in young infants. • Features of fracture healing develop in a logical progression. • Evidence provided for determining fractures are consistent with a proposed time frame. • It is of critical importance to have sound evidence for the dating of fractures.

Healing patterns of clavicular birth injuries as a guide to fracture dating in cases of possible infant abuse

BACKGROUND

Dating fractures is critical in cases of suspected infant abuse. There are little scientific data to guide radiologists, and dating is generally based on personal experience and conventional wisdom.

OBJECTIVE

Since birth-related clavicular fractures are not immobilized and their age is known, we propose that an assessment of these injuries may serve as a guide for dating inflicted fractures in young infants, acknowledging that patterns observed in the clavicle may not be entirely generalizable to other bones injured in the setting of abuse.

MATERIALS AND METHODS

One hundred thirty-one radiographs of presumed birth-related clavicular fractures in infants between 0 and 3 months of age were reviewed by two pediatric radiologists with 30 and 15 years' experience. Readers were asked to evaluate images based on several parameters of fracture healing, with a focus on subperiosteal new bone formation (SPNBF) and callus formation. SPNBF and callus were each evaluated with regard to presence, thickness and character. Responses were correlated with known fracture ages.

RESULTS

SPNBF was rarely seen in fractures less than 7 days old and was most often present by 10 days. Callus formation was rarely seen in fractures less than 9 days old and was most often present by 15 days. SPNBF thickness increased with fracture age and the character of SPNBF evolved from single-layered to solid/multilayered. Callus thickness decreased with fracture age and callus matrix evolved from soft to intermediate to hard in character.

CONCLUSION

There is an evolution in clavicular fracture healing in young infants that follows a predictable pattern. These findings afford the prospect that predictable patterns of infant clavicular fracture healing can provide an evidence base that may be applicable in cases of suspected infant abuse.