%0 Conference Paper %1 MedIm2016 %A Pfitzner, C. %A May, S. %A Nüchter, A. %B Proceedings of the SPIE 9784, Medical Imaging 2016: Image Processing %C San Diego, CA, USA %D 2016 %K imported %R 10.1117/12.2216042 %T Neural Network-based Visual Body Weight Estimation for Drug Dosage Finding %X Body weight adapted drug dosages are important for emergency treatments. This paper describes an improved body weight estimation approach for emergency patients in a trauma room, based on images from a RGBD sensor and a thermal camera. The improvements are archived by several extensions: The sensor fusion of RGBD and thermal camera eases filtering and segmentation of the patient's body from the background. Robustness and accuracy is gained by an artificial neural network (ANN), which considers features from the sensors as input to calculate the patient's body weight, e.g. the patient's volume, surface and shape parameters. The ANN is trained offline with 30 percent of the patients data. Preliminary experiments with 69 real patients show an accuracy close to 90 percent for a threshold of ten percent relative error in real body estimation. Results are compared to the patient's self estimation, a physician's guess and an anthropometric method: If the patient is knowledgeable it is the best possibility for body weight adapted drug dosages with 97 percent accuracy. The treating physicians and the anthropometric estimation achieve an accuracy of approximately 70 percent. The here presented approach gets an accuracy of nearly 90 percent and would be the best solution if a patient can not provide his own body weight and can not be weighted on a scale. These preliminary results demonstrate a sufficient approach for an upcoming clinical trial with 1,000 patients for body weight estimation.

@inproceedings{MedIm2016, abstract = {Body weight adapted drug dosages are important for emergency treatments. This paper describes an improved body weight estimation approach for emergency patients in a trauma room, based on images from a RGBD sensor and a thermal camera. The improvements are archived by several extensions: The sensor fusion of RGBD and thermal camera eases filtering and segmentation of the patient's body from the background. Robustness and accuracy is gained by an artificial neural network (ANN), which considers features from the sensors as input to calculate the patient's body weight, e.g. the patient's volume, surface and shape parameters. The ANN is trained offline with 30 percent of the patients data. Preliminary experiments with 69 real patients show an accuracy close to 90 percent for a threshold of ten percent relative error in real body estimation. Results are compared to the patient's self estimation, a physician's guess and an anthropometric method: If the patient is knowledgeable it is the best possibility for body weight adapted drug dosages with 97 percent accuracy. The treating physicians and the anthropometric estimation achieve an accuracy of approximately 70 percent. The here presented approach gets an accuracy of nearly 90 percent and would be the best solution if a patient can not provide his own body weight and can not be weighted on a scale. These preliminary results demonstrate a sufficient approach for an upcoming clinical trial with 1,000 patients for body weight estimation. }, added-at = {2017-09-19T13:40:53.000+0200}, address = {San Diego, CA, USA}, author = {Pfitzner, C. and May, S. and N{\"u}chter, A.}, biburl = {https://www.bibsonomy.org/bibtex/2a3c3662cd2f6fe332f7471fe6dc44c3c/nuechter76}, booktitle = {Proceedings of the SPIE 9784, Medical Imaging 2016: Image Processing}, doi = {10.1117/12.2216042}, interhash = {8d7cd8a8fc62c651e3fe6778e2dd16c3}, intrahash = {a3c3662cd2f6fe332f7471fe6dc44c3c}, keywords = {imported}, month = {February}, timestamp = {2017-12-21T22:01:39.000+0100}, title = {Neural Network-based Visual Body Weight Estimation for Drug Dosage Finding}, year = 2016 }