Burn-injured tissue detection for debridement surgery through the combination of non-invasive optical imaging techniques

Review of machine learning for optical imaging of burn wound severity assessment

Significance

Over the past decade, machine learning (ML) algorithms have rapidly become much more widespread for numerous biomedical applications, including the diagnosis and categorization of disease and injury.

Aim

Here, we seek to characterize the recent growth of ML techniques that use imaging data to classify burn wound severity and report on the accuracies of different approaches.

Approach

To this end, we present a comprehensive literature review of preclinical and clinical studies using ML techniques to classify the severity of burn wounds.

Results

The majority of these reports used digital color photographs as input data to the classification algorithms, but recently there has been an increasing prevalence of the use of ML approaches using input data from more advanced optical imaging modalities (e.g., multispectral and hyperspectral imaging, optical coherence tomography), in addition to multimodal techniques. The classification accuracy of the different methods is reported; it typically ranges from ∼ 70 % to 90% relative to the current gold standard of clinical judgment.

Conclusions

The field would benefit from systematic analysis of the effects of different input data modalities, training/testing sets, and ML classifiers on the reported accuracy. Despite this current limitation, ML-based algorithms show significant promise for assisting in objectively classifying burn wound severity.

Non-invasive medical imaging technology for the diagnosis of burn depth

Currently, the clinical diagnosis of burn depth primarily relies on physicians' judgements based on patients' symptoms and physical signs, particularly the morphological characteristics of the wound. This method highly depends on individual doctors' clinical experience, proving challenging for less experienced or primary care physicians, with results often varying from one practitioner to another. Therefore, scholars have been exploring an objective and quantitative auxiliary examination technique to enhance the accuracy and consistency of burn depth diagnosis. Non-invasive medical imaging technology, with its significant advantages in examining tissue surface morphology, blood flow in deep and changes in structure and composition, has become a hot topic in burn diagnostic technology research in recent years. This paper reviews various non-invasive medical imaging technologies that have shown potential in burn depth diagnosis. These technologies are summarized and synthesized in terms of imaging principles, current research status, advantages and limitations, aiming to provide a reference for clinical application or research for burn specialists.

Lighting the Way for Necrosis Excision Through Indocyanine Green Fluorescence-Guided Surgery

BACKGROUND

No objective technique exists to distinguish necrotic from viable tissue, risking over-excision in burns and loss of wound healing potential. Second window indocyanine green (SWIG) is a novel fluorescence-imaging modality being studied to identify residual solid tumors during oncological surgery. SWIG has also been shown to have avidity for necrosis in animal models, but translation of these findings to humans is lacking. The objective of this study was to evaluate SWIG in the identification of burn wound necrosis and compare it with previously published indocyanine green angiography (ICGA) techniques.

STUDY DESIGN

This study used mouse, human skin xenograft and human patient burn models. Brightfield and SWIG near-infrared imaging were performed on macroscopic tissue samples, which were then cryopreserved, sectioned, and analyzed for microscopic fluorescence. SWIG fluorescence findings were correlated to visual assessment of the burn wound as well as histological markers of necrosis using hematoxylin and eosin and lactate dehydrogenase stains.

RESULTS

We found that SWIG identified burn necrosis in a manner dependent on the dose and timing of indocyanine green (ICG) administration and had an inverse fluorescence signal compared with ICGA. Furthermore, SWIG fluorescence identified the interface of viable and nonviable tissue.

CONCLUSION

Our study confirmed that ICGA is an inconsistent and nonstandardized modality to evaluate burn injuries. In contrast, SWIG imaging is a potential imaging modality to objectively prognosticate burn wound healing potential and guide intraoperative burn excision. Further studies are needed to define ratios of fluorescence intensity values to guide surgical decision-making in burn excision and to better define how ICG is retained in necrotic tissue to enhance utility of SWIG in other disease processes.

Evaluation of hyperspectral imaging as a modern aid in clinical assessment of burn wounds of the upper extremity

The most common burn wound assessment continues to be the clinical inspection and the tactile examination, which are subjective and remain challenging even for experienced burn surgeons. Recently, hyperspectral imaging camera systems have been increasingly used to support the evaluation of burn wounds. The aim of our study was to determine if hyperspectral imaging analysis differentiates and objectifies the assessment of burn wounds in burns of the upper extremities. We included 97 superficial partial, deep partial dermal burns, and full thickness burns. Hyperspectral imaging analysis was performed for all burns using proprietary software. The software recorded parameters for tissue oxygenation (StO2), tissue hemoglobin index, and near-infrared perfusion. These values were compared with the recordings for healthy, non-burned skin. We found that hyperspectral imaging analysis effectively differentiates burn wounds and shows the ability to distinguish even superficial partial burns from deep partial burns in the near-infrared perfusion analysis feature. Although, it was not possible to differentiate burn wounds in all features. Currently, it is important to optimize the respective reference values of the individual burn degrees for an objectified assessment.

Artificial intelligence in the management and treatment of burns: a systematic review

BACKGROUND

Artificial intelligence (AI) is an innovative field with potential for improving burn care. This article provides an updated review on machine learning in burn care and discusses future challenges and the role of healthcare professionals in the successful implementation of AI technologies.

METHODS

A systematic search was carried out on MEDLINE, Embase and PubMed databases for English-language articles studying machine learning in burns. Articles were reviewed quantitatively and qualitatively for clinical applications, key features, algorithms, outcomes and validation methods.

RESULTS

A total of 46 observational studies were included for review. Assessment of burn depth (n = 26), support vector machines (n = 19) and 10-fold cross-validation (n = 11) were the most common application, algorithm and validation tool used, respectively.

CONCLUSION

AI should be incorporated into clinical practice as an adjunct to the experienced burns provider once direct comparative analysis to current gold standards outlining its benefits and risks have been studied. Future considerations must include the development of a burn-specific common framework. Authors should use common validation tools to allow for effective comparisons. Level I/II evidence is required to produce robust proof about clinical and economic impacts.

First preliminary Clinical Experiences using Hyperspectral Imaging for Burn Depth Assessment of Hand Burns

The accurate assessment of burn wounds is challenging but crucial for correct diagnosis and following therapy. The most frequent technique to evaluate burn wounds remains the clinical assessment often subjective depending on the experience of the physician. Hyperspectral Imaging is intended to counteract this subjective diagnosis by an accurate and objective analysis of perfusion parameters. The purpose of this study was to analyse the ability of technical burn depth assessment and to investigate a possible link between a certain value to burn depth versus value of healthy skin references.

METHODS

A total of 118 subjects were included in this study between July 2017 and July 2019. 74 images with dorsal hand burns and 44 images of healthy skin on the dorsal hand as control group were analysed. In Hyperspectral Imaging recordings burn wounds were analysed with special interest to wound centre, intermediate zone, and wound margin.

RESULTS

Significant results were determined for the differentiation between superficial partial burns and healthy skin. Furthermore, the distinction of full thickness burns was significantly possible.

CONCLUSION

Currently, it cannot be shown that the use of Hyperspectral Imaging technology significantly assesses the actual burn depth of thermal wounds of the dorsal hand reliably. However, the results show tendencies to improved analysis for differentiations supporting physicians in early objective optimal treatment selection.

Towards quantitative assessment of burn based on photoacoustic and optical coherence tomography

Accurate and timely assessment of the severity of burn is essential for the treatment of burns. Currently, although most first-degree and third-degree burns are easily diagnosed through visual inspection or auxiliary diagnostic methods, the second-degree burn is still difficult to distinguish due to the ambiguity boundaries of second-degree with first-degree and third-degree burns. In this study, we proposed a non-invasive technique by combing photoacoustic imaging (PAI) and optical coherence tomography (OCT) to multi-parameter quantitatively assess the burns. The feasibility and capacity of the dual-mode PAT/OCT for assessing the burns was first testified by tissue-mimicking phantom and burn wounds in mouse pinna in vivo. The further experiments conducted on the back of rats showed that the changes in skin scattering structure, vascular morphology and blood flow provided by the dual-mode PAI/OCT system can determine distinct boundaries and depth of the burns. The experimental results prove that combined PAI/OCT as a novel method can be used to assess the severity of burn, which has the potential to diagnose the burns in clinic.

Initial Experience With Autologous Skin Cell Suspension for Treatment of Deep Partial-Thickness Facial Burns

Facial burns present a challenge in burn care, as hypertrophic scarring and dyspigmentation can interfere with patients’ personal identities, ocular and oral functional outcomes, and have long-term deleterious effects. The purpose of this study is to evaluate our initial experience with non-cultured, autologous skin cell suspension (ASCS) for the treatment of deep partial-thickness (DPT) facial burns. Patients were enrolled at a single burn center during a multicenter, prospective, single-arm, observational study involving the compassionate use of ASCS for the treatment of large total BSA (TBSA) burns. Treatment decisions concerning facial burns were made by the senior author. Facial burns were initially excised and treated with allograft. The timing of ASCS application was influenced by an individual’s clinical status; however, all patients were treated within 30 days of injury. Outcomes included subjective cosmetic parameters and the number of reoperations within 3 months. Five patients (4 males, 1 female) were treated with ASCS for DPT facial burns. Age ranged from 2.1 to 40.7 years (mean 18.2 ± 17.3 years). Average follow-up was 231.2 ± 173.1 days (range 63–424 days). Two patients required reoperation for partial graft loss within 3 months in areas of full-thickness injury. There were no major complications and one superficial hematoma. Healing and cosmetic outcomes were equivalent to, and sometimes substantially better than, outcomes typical of split-thickness autografting. Non-cultured, ASCS was successfully used to treat DPT facial burns containing confluent dermis with remarkable cosmetic outcomes. Treatment of DPT burns with ASCS may be an alternative to current treatments, particularly in patients prone to dyspigmentation, scarring sequelae, and with limited donor sites.

Engineering Tissue Fabrication With Machine Intelligence: Generating a Blueprint for Regeneration

Regenerating lost or damaged tissue is the primary goal of Tissue Engineering. 3D bioprinting technologies have been widely applied in many research areas of tissue regeneration and disease modeling with unprecedented spatial resolution and tissue-like complexity. However, the extraction of tissue architecture and the generation of high-resolution blueprints are challenging tasks for tissue regeneration. Traditionally, such spatial information is obtained from a collection of microscopic images and then combined together to visualize regions of interest. To fabricate such engineered tissues, rendered microscopic images are transformed to code to inform a 3D bioprinting process. If this process is augmented with data-driven approaches and streamlined with machine intelligence, identification of an optimal blueprint can become an achievable task for functional tissue regeneration. In this review, our perspective is guided by an emerging paradigm to generate a blueprint for regeneration with machine intelligence. First, we reviewed recent articles with respect to our perspective for machine intelligence-driven information retrieval and fabrication. After briefly introducing recent trends in information retrieval methods from publicly available data, our discussion is focused on recent works that use machine intelligence to discover tissue architectures from imaging and spectral data. Then, our focus is on utilizing optimization approaches to increase print fidelity and enhance biomimicry with machine learning (ML) strategies to acquire a blueprint ready for 3D bioprinting.

Burn wound classification model using spatial frequency-domain imaging and machine learning

Accurate assessment of burn severity is critical for wound care and the course of treatment. Delays in classification translate to delays in burn management, increasing the risk of scarring and infection. To this end, numerous imaging techniques have been used to examine tissue properties to infer burn severity. Spatial frequency-domain imaging (SFDI) has also been used to characterize burns based on the relationships between histologic observations and changes in tissue properties. Recently, machine learning has been used to classify burns by combining optical features from multispectral or hyperspectral imaging. Rather than employ models of light propagation to deduce tissue optical properties, we investigated the feasibility of using SFDI reflectance data at multiple spatial frequencies, with a support vector machine (SVM) classifier, to predict severity in a porcine model of graded burns. Calibrated reflectance images were collected using SFDI at eight wavelengths (471 to 851 nm) and five spatial frequencies (0 to 0.2  mm  -  1). Three models were built from subsets of this initial dataset. The first subset included data taken at all wavelengths with the planar (0  mm  -  1) spatial frequency, the second comprised data at all wavelengths and spatial frequencies, and the third used all collected data at values relative to unburned tissue. These data subsets were used to train and test cubic SVM models, and compared against burn status 28 days after injury. Model accuracy was established through leave-one-out cross-validation testing. The model based on images obtained at all wavelengths and spatial frequencies predicted burn severity at 24 h with 92.5% accuracy. The model composed of all values relative to unburned skin was 94.4% accurate. By comparison, the model that employed only planar illumination was 88.8% accurate. This investigation suggests that the combination of SFDI with machine learning has potential for accurately predicting burn severity.

Mahalanobis Outier Removal for Improving the Non-Viable Detection on Human Injuries

Machine learning techniques have been recently applied for discriminating between Viable and Non-Viable tissues in animal wounds, to help surgeons to identify areas that need to be excised in the process of burn debridement. However, the presence of outliers in the training data set can degrade the performance of that discrimination. This paper presents an outlier removal technique based on the Mahalanobis distance to improve the accuracy detection of Non-Viable skin in human injuries. The iteratively application of this technique improves the accuracy results of the Non-Viable skin in a 13.6% when applying K-fold cross-validation.

Evaluating clinical observation versus Spatial Frequency Domain Imaging (SFDI), Laser Speckle Imaging (LSI) and thermal imaging for the assessment of burn depth

While clinical examination is needed for burn severity diagnosis, several emerging technologies aim to quantify this process for added objectivity. Accurate assessments become easier after burn progression, but earlier assessments of partial thickness burn depth could lead to earlier excision and grafting and subsequent improved healing times, reduced rates of scarring/infection, and shorter hospital stays. Spatial Frequency Domain Imaging (SFDI), Laser Speckle Imaging (LSI) and thermal imaging are three non-invasive imaging modalities that have some diagnostic ability for noninvasive assessment of burn severity, but have not been compared in a controlled experiment. Here we tested the ability of these imaging techniques to assess the severity of histologically confirmed graded burns in a swine model. Controlled, graded burn wounds, 3cm in diameter were created on the dorsum of Yorkshire pigs (n=3, 45-55kg) using a custom-made burn tool that ensures consistent pressure has been employed by various burn research groups. For each pig, a total of 16 burn wounds were created on the dorsal side. Biopsies were taken for histological analysis to verify the severity of the burn. Clinical analysis, SFDI, LSI and thermal imaging were performed at 24 and 72h after burn to assess the accuracy of each imaging technique. In terms of diagnostic accuracy, using histology as a reference, SFDI (85%) and clinical analysis (83%) performed significantly better that LSI (75%) and thermography (73%) 24h after the burn. There was no statistically significant improvement from 24 to 72h across the different imaging modalities. These data indicate that these imaging modalities, and specifically SFDI, can be added to the burn clinicians' toolbox to aid in early assessment of burn severity.

Comparison of speckleplethysmographic (SPG) and photoplethysmographic (PPG) imaging by Monte Carlo simulations and in vivo measurements

Noncontact photoplethysmography (PPG) is limited by a poor signal-to-noise ratio (SNR). A solution to this limitation is the use of alternate sources of optical contrast to generate a complementary pulsatile waveform. One such source is laser speckle contrast, which is modulated in biological tissues by the flow rate of red blood cells. Averaging a region of interest from a speckle contrast image over time allows for the calculation of a speckleplethysmogram (SPG). Similar to PPG, SPG enables monitoring of heart rate and respiratory rate. A gap in the knowledge base exists as to the precise spatiotemporal relationship between PPG and SPG signals. We have developed an eight-layer tissue model to simulate both PPG and SPG signals in a reflectance geometry via Monte Carlo methods. We modeled PPG by compression of the upper and lower blood nets due to expansion of the larger arterial layer below. The in silico PPG peak-to-peak amplitude percent was greater at 532 nm than at 860 nm (5.6% vs. 3.0%, respectively), which matches trends from the literature. We modeled SPG by changing flow speeds of red blood cells in both the capillaries and arterioles over the cardiac cycle. The in silico SPG peak-to-peak amplitude percent was 24% at 532 nm and 40% at 860 nm. In silico results are similar to in vivo results measured with a two-camera set up for simultaneous imaging of PPG and SPG. Both in silico and in vivo data suggest SPG has a much larger SNR than PPG, which may prove beneficial for noncontact, wide-field optical monitoring of cardiovascular health.

Merging of Classifiers for Enhancing Viable vs Non-Viable Tissue Discrimination on Human Injuries

Non-invasive optical imaging techniques have been recently proposed for distinguishing between different types of tissue in burns generated in porcine models. These techniques are designed to assist surgeons during the process of burn debridement, to identify regions requiring excision and their appropriate excision depth. This paper presents a machine learning tool for discriminating between Viable and Non- Viable tissues in human injuries. This tool merges a supervised (QDA) with an unsupervised (k-means clustering) classification algorithms. This combination improves the Non-Viable tissue detection in 23.7% with respect to a simple QDA classifier.