Key Points
- Researchers from the Nestlé Institute of Agricultural Sciences developed a rapid, scalable, and cost-effective method using reflectance spectroscopy and machine learning to detect cadmium levels in root vegetables as an alternative to traditional laboratory methods.
- The study achieved high prediction accuracy, with regression models reaching R² values up to 0.95 and classification models attaining F1-scores over 0.96.
- Key takeaways include the identification of crucial wavelengths for cadmium detection and the model’s robustness even at thresholds five times stricter than current regulations.
Trace metal elements (TMEs) are a chief concern in the food industry. One example of a TME is cadmium, which is a toxic heavy metal that can be found in several key food products. An ongoing challenge in the food industry is that traditional laboratory methods used to detect TME content in food are expensive, time-consuming processes.
A recent study published in Food Control explored this issue. In the study, researchers from the Nestlé Institute of Agricultural Sciences in France introduced a new, rapid, and scalable method to detect cadmium in root vegetables using reflectance spectroscopy and machine learning (1). The results indicate that their method has the potential to be a cost-effective alternative.
Why is cadmium toxic in food?
Cadmium intake is toxic to humans because of its effect on the human body. Although removing cadmium completely from food is impossible, the amount of cadmium has to be carefully monitored because when ingested, cadmium can result in humans experiencing abdominal pain and cramps, nausea, vomiting, and diarrhea (2). Prolonged exposure to cadmium has even more deleterious effects, including respiratory illnesses, cardiovascular disease, diabetes, and bone demineralization (2).
Currently, the international regulations provide strict guidelines for cadmium content in food. For example, European Union caps cadmium levels in fresh carrots at 0.10 mg per kilogram of fresh weight. However, current methods used to ensure regulatory compliance—such as atomic absorption spectroscopy or inductively coupled plasma mass spectrometry—are expensive, time-consuming, and often not scalable for large agricultural systems.
Cadmium in particularly a concern in agricultural produce such as root vegetables (1). This is because cadmium is often present in places where phosphate fertilizers are used, and this includes farming and industrial areas where fossil fuels are burned (2).
What was the method the researchers tested in their study?
In their study, the researchers investigated cadmium levels in carrots using reflectance spectroscopy. Reflectance spectroscopy is an optical technique that is emerging in several application areas (3). In the study, the researchers used reflectance spectroscopy to analyze how carrot tissues reflect light across the 350–2500 nm spectral range (1). The researchers built machine learning (ML) models capable of either estimating the actual cadmium concentration (regression models) or determining whether the concentration exceeds regulatory limits (classification models) (1).
What were the results of the study?
The researchers found that their regression and classification models based on cross-sections of carrot roots demonstrated prediction accuracy as high as 95%, with R² values of 0.95 and F1-scores exceeding 0.96 (1).
The researchers also examined the efficacy of less invasive sampling techniques by investigating models based on leaf reflectance and unpeeled roots (1). Although these approaches were less accurate than those using cross-sections, they still showed moderate to high predictive power, with R² values between 0.48 and 0.87 and F1-scores ranging from 64 to 90 (1).
Six models were built in this study, with three being regression models and three being classification models. These models used the spectral data collected from three carrot tissue types: leaves (indirect model), peels (non-destructive model), and cross-sections (destructive model) (1). By building all these models, the researchers were able to compare their performance across more tissue types (1).
What are the key takeaways from the study?
One of the key takeaways from the study was the identification of specific wavelengths in the visible and short-wave infrared (IR) regions that contributed most significantly to accurate predictions. Another key takeaway is that this study tested the models against cadmium thresholds five times lower than the current regulatory limit, and the models maintained their reliability (1).
However, these models could be further improved. The researchers acknowledged that future studies should focus on validating these models in field conditions and expanding their use to other trace metals and crops (1).
References
- Maugeais, N.; Lassalle, G. Predicting Cadmium Accumulation in Carrot (Daucus carota L.) Using Reflectance Spectroscopy and Machine Learning. Food Cont. 2025, 173, 11226. DOI: 10.1016/j.foodcont.2025.111226
- U.S. Food and Drug Administration, Cadmium in Food and Foodwares. U.S. Food and Drug Administration. Available at: https://www.fda.gov/food/environmental-contaminants-food/cadmium-food-and-foodwares#:~:text=Health%20Effects%20Information,Control%20and%20Prevention%20(CDC). (accessed 2025-06-30).
- Wallace, M. B.; Wax, A.; Roberts, D. N.; Graf, R. N. Reflectance Spectroscopy. Gastrointest. Endosc. Clin. N. Am. 2009, 19 (2), 233–242. DOI: 10.1016/j.giec.2009.02.008
