Hologram Sciences employs AI tech to improve patient nutritional monitoring

12 Feb 2024 --- Hologram Sciences will integrate its Nutrition Multimodal Language Model (NuMLM) in healthcare settings to provide accurate, real-time nutritional monitoring at scale. The company is preparing for the first clinical application of the model — a key component of its Precision Nutrition Platform — at the US Mayo Clinic, aiming to address malnutrition challenges in surgical recovery and enhance patient care.

“NuMLM is one of the first multimodal language models specifically trained for advanced nutrition tracking and malnutrition screening in clinical settings,” Ian Brady, CEO of Hologram Sciences, tells Nutrition Insight. “This allows us to address a critical gap in patient care given that 42% of hospitalized adult patients have some level of malnourishment, leading to longer stays and higher readmission rates.”

“Accurately tracking nutrition for every patient is practically impossible because it is both operationally intense and manual, so prone to human error. NuMLM applies AI to a clinical need that will dramatically benefit from the speed, scale and accuracy only this technology can provide.”

He underscores that the specific training data used on the model outperforms generic models in accuracy but also changes how nutritional intake is assessed, “moving from a process that is typically untracked or inaccurate, to an exact, scalable approach.”

Through a know-how agreement, NuMLM will draw on the Mayo Clinic’s clinical insights. Brady adds that the company will work closely with Mayo Clinic to validate clinical impact. “A specific timeline has not been set as we focus on meeting the highest standards of accuracy for clinical support and adoption.”

How does it work?
Brady explains that NuMLM is designed to be simple to implement in a clinical setting. Through pre- and post-consumption images of patient meals, it automatically measures what nutrients they consumed.

“The models are trained on a proprietary database of food images and precise nutritional measurements that Hologram Sciences has assembled over many years with advanced image segmentation techniques. It also employs registered dietitians to validate nutritional assessments, which further trains the model and ensures the accuracy of electronic health records.”

The multimodal foundation model is trained on Hologram Sciences’ food-image database and is part of the company’s Precision Nutrition Platform, on which it also cooperates with the Mayo Clinic.

This image database contains high-quality historical images collected across years of meal tracking, explains Brady. “Each image has been assessed for precise nutritional values such as calories, macronutrients and micronutrients.”

“This specialized set of training data enables exact recognition and nutritional evaluation of a wide range of food items and portion sizes. Advanced image segmentation techniques are used to analyze the quantity of food before and after consumption, vital for precise calculations of patients’ nutritional intake.”

Improving patient care
According to Brady, the technology improves patient care across inpatient, outpatient and post-surgery settings across various medical disciplines. “NuMLM will automate tracking of a patient’s macro and micronutrient intake post-consumption so that clinical staff can make timely interventions.”

“Take wound healing, for example. This is a common concern in surgical recovery, where nutrition is vital. A patient needs a steady intake of nutrients like protein, vitamin C and zinc throughout the day to support optimal healing. They must also meet their overall calorie needs because wound healing increases energy demand.”

NuMLM measures consumed nutrients by comparing pre- and post-consumption images of patient meals (Image credit: Hologram Sciences).“Currently, patient nutritional intake is measured by highly manual, error-prone processes,” he illustrates. “Typically, a meal is ordered for or by a patient with a known calorie and nutrient profile. After consuming the meal, a nurse removing a food tray from a patient’s room will estimate what percentage of the food was consumed and enter relevant notes into the patient record, sometimes hours later. Then a registered dietitian will reference those notes and manually calculate nutrients.”

He cautions that this process is time-consuming, which may result in not recording nutritional intake for all patients as care focuses primarily on the highest-risk patients. This leaves the risks of malnutrition across the entire patient population largely unmanaged.

“NuMLM can provide a superior approach given its ability to take precise nutritional measurements automatically, in real-time, for every single patient and better assess risk across all patients,” emphasizes Brady.

Additional benefits
NuMLM will also be integrated into Hologram Sciences’ more extensive Precision Nutrition Platform. Brady adds that this platform can identify potential health risks from dietary patterns and provide nutritional interventions depending on patients’ needs.

Moreover, since the technology will be employed consistently during a patient’s hospital stay, it can recognize the cumulative impact of subtle patterns that an individual clinical staff member might not detect from seeing one meal.

“For example, the system could identify that a patient is getting too many of her calories from carbs and not enough protein or fiber or is deficient in particular nutrients that support wound healing,” details Brady.

He concludes: “This AI-driven approach allows for personalized nutrition management at scale, taking into account specific diagnostics and clinical protocols, ensuring that each patient gets tailored support to improve their outcomes.”

By Jolanda van Hal

To contact our editorial team please email us at editorial@cnsmedia.com