Research Series: Article 3
August 20, 2019
Shaping the Future of Patient Care with Artificial Intelligence (AI)
By Raphael Yaakov, MS and Kyle Wu, MD
There has never been a more exciting time than now to be at the intersection of health and technology. The breakthroughs in science and technology are paving the path for personalized precision care. In wound care, especially, we have experienced tremendous growth. Simple gauze dressings have evolved to biomaterials that mimic the native extracellular matrix, ultrasound technology revolutionized debridement, and autograft and allograft preparations gave way to tissue-engineered skin replacement. Our understanding of the wound microenvironment and healing has deepened. It has truly been an age of progress, but the gains have not been equitable across all domains of wound care.
Despite the proliferation of AI-powered mobile applications, little attention has been given to wound measurement and assessment practices. In clinical practice, the ruler method continues to be a mainstream protocol. The current wound assessment uses limited descriptors and is dependent on the observational skills, knowledge and judgment of a nurse or clinician. Furthermore, complex wounds with mixed etiologies can make identification and diagnosis difficult. With increasing regulations and demands to deliver improved clinical outcomes, it is becoming a necessity for wound care specialists to effectively adopt technology into their practice. AI presents a plethora of advantages over traditional clinical decision-making. Essentially, AI will give clinicians unprecedented insights to enable early diagnosis, timely and personalized intervention, and ultimately improve patient outcomes and reduce costs.
The game-changing innovations in computing has enabled the deep learning system (DLS) to learn from inputs (or examples) rather than rules defined in code. In a nutshell, deep learning is analogous to our biological neural network which receives inputs from other neurons and delivers the activation signal to the next layer in the network when a threshold value is reached. Similarly, neural network algorithms use data from each previous layer to inform results; they can be trained to fire when it detects a high probability of a feature related to a desired prediction. The convolutional neural network (CNN) is inspired by the visual cortex; it detects simple representations, such as edges, lines and curves in initial layers to more complex representations of an image or object in the final layers.
The power of deep learning is unleashed as more data is collected and analyzed. In a noteworthy study by Ting et al, over 200,000 retinal images were used to train and inform a CNN on identification of diabetic retinopathy and related eye diseases.1 The model was comparable with trained graders in detecting referable diabetic retinopathy and had a higher sensitivity for detecting vision-threatening diabetic retinopathy, however specificity was lower than trained graders.1 In a more recent study, Guler and colleagues offered a novel hybrid model consisting of deep convolutional network (DCN) and deep neural network (DNN) to identify wound etiology using both wound images and patient demographics. The model achieved an accuracy of 94% in differentiating diabetic, venous and burn wounds.2 The team also proposed a DLS to predict wound healing using long short-term memory.3 These early works showcase the potential of deep learning.
While AI holds promise for reducing human error, it is not immune to built-in nuances, biases and errors. Thus, it is imperative for the model to maintain the flexibility to integrate new information and input from clinicians. If the training dataset is inadequate, the system will fail to produce the desired outcome. Data is the foundation of AI. Lack of curated datasets can seriously undermine the efficacy of DLS. Adding to the problem is the issue that there is currently a lack of standardized guidelines on development and implementation of AI applications in healthcare. AI technology is evolving rapidly, and the FDA is playing catch-up. AI presents a unique challenge as it does not neatly fit the current regulatory pathway for software as a medical device (SaMD). Complex DLS has multiple hidden layers of decision-making which can be challenging to evidence. Moreover, the inherent continuous learning algorithm in AI applications may change and adapt with a given set of inputs, ultimately producing different outputs and impacting intended use. Embracing and advancing AI will require effective collaboration between clinicians, data scientists, programmers, and regulators to identify and control the risk associated with development and the ongoing use of AI applications; education and understanding of AI is the first step towards this partnership.
Raphal Yaakov’s experience spans across phase I-IV drug and device multinational trials. He currently serves as Vice President of Clinical Operations at SerenaGroup, a Collaborative group dedicated to wound care research.
Dr. Kyle Wu is Chief Medical Officer and co-founder of eKare. Dr. Wu received surgical training at the Georgetown University Hospital and MD/MBA degrees from Columbia University. He has numerous peer‐reviewed publications in the fields of optics, imaging, and robotic surgery.
The views and opinions expressed here are those of the author and do not necessarily reflect the official policy or position of any other agency, organization, employer or company.
1Ting DSW, Cheung CY, Lim G, et al. Development and Validation of a Deep Learning System for Diabetic Retinopathy and Related Eye Diseases Using Retinal Images From Multiethnic Populations With Diabetes. JAMA. 2017;318(22):2211–2223. doi:10.1001/jama.2017.18152.
2Guler O, Wu K, Cheng P, et al. Using Artificial Intelligence (AI) to Identify Wound Etiology: A Preliminary Study. Innovations in Wound Healing (IWH) Conference. Key West, Florida. 2018.
3Guler O, Cheng P, Wu K. Using Artificial Intelligence (AI) to Model Healing Prediction: A Preliminary Study. Symposium on Advanced Wound Care. San Antonio, TX. 2019.