NEC Thermography Infrared Imaging Study

Background information about NEC:

The best prognosis requires early recognition and treatment of NEC. Perforation or ongoing deterioration may require surgical intervention. Mortality has improved in recent years but is still about 15-30 % for non-surgical NEC and up to 50% for surgical NEC. Survivors may suffer significant complications including long-term damage to bowel function.

The initial diagnosis is based on clinical exam and diagnostic imaging, usually involving abdominal x-rays +/- ultrasound. Bell's staging criteria incorporates clinical, laboratory, and radiographic findings, and is used to discern suspected NEC from definitive NEC. Bell's stage 2 denotes proven NEC and involves the presence of pneumatosis intestinalis and/or portal venous gas on x-ray.

The main difficulty when diagnosing NEC is that early/suspected stages of NEC present clinically and radiographically in a similar way to benign feeding intolerance seen with prematurity. Abdominal x-rays involve exposure of the patient to ionizing radiation and ultrasound imaging is disruptive to the patient, takes time to perform, and requires skilled operators who have limited availability. These modalities will only show non-specific changes in early stages of NEC. Late findings are more discerning, however by the time these findings are apparent the optimal time frame to initiate treatment is often missed.

Background Information About Medical Thermography:

Medical infrared (IR) imaging is a combination of medical technology, infrared camera technology and computer multimedia technology. Infrared imaging devices record the human thermal fields. The human body is a natural biological heating source and an infrared thermal imager converts the far-infrared light wave radiated by the human body to an electrical signal which is then converted into a digital quantity by analog/digital conversion. The signals are processed by multimedia image processing technology into a color heat map, showing the body's temperature field. A normal, healthy body has a characteristic heat map while an abnormal body exhibits deviations from this heat map. Thus, comparing the similarities and differences between the two, in combination with clinical diagnosis, doctors and researchers can infer the nature and extent of a disease.

The investigators investigated various methods of computerized assessment of thermal images obtained from premature babies with definitive NEC compared with normal babies with no symptoms of feeding intolerance. The investigators showed that the discriminating power of the surface temperature evolution data between healthy control babies and those who were diagnosed with NEC was promising, with fairly high classification rate and a simple linear classification scheme. In addition, our results suggested a slower rate of decrease of abdominal skin temperature in the NEC babies, which might be explained by the presence of inflamed bowel. Rice et al. also used medical thermography to help detect NEC in newborn infants admitted to the NICU. By comparing temperature distribution over abdominal segments to those of the chest, they were able to determine differences in normal babies vs those diagnosed with NEC.

Background Information About Medical Microsoft Kinect Sensor:

The Kinect sensor is a widely used, consumer-grade, and safe color and depth camera commercially marketed for use in computer gaming. It measures near infrared light reflected off the body to form a surface map of the environment. The RGB-D Kinect sensor has recently been applied to healthcare in a variety of applications where it has shown to be effective, including the field of medical rehabilitation, where Kinect sensors are used for active exercise training and rehabilitation of patients,for safe and automated radiotherapy delivery, for chest wall motion analysis in cystic fibrosis patients, for facial feature analysis of posture control in rehabilitation patients, in ICU patients for automated mobility measurements, and even automated apnea monitoring in infants and children.

In this study, the RGB-D Kinect sensor's role will essentially consist of supporting the automated segmentation, and retrieval of thermal distribution data from the co-located and calibrated IR FLIR imager. Thanks to the color and depth maps that will be made available, classical and robust image processing techniques will be leveraged to identify regions of interest (the baby's body and eventually abdomen). This sort of precise segmentation cannot be reliably performed solely from IR images, as it is well established that computer vision and image processing techniques are not well adapted for IR images. Early laboratory tests have already been conducted to determine that the operation of the RGB-D sensor in an overlapping field of view with that of the IR imager does not influence the data collected in the heat map. The combined operation is therefore safe and will provide an accurate thermal distribution as if the IR imager was operating alone. The end goal is to combine the strengths of three imaging technologies to improve medical diagnosis.

Study Rationale:

Previous studies investigating thermography and diagnosing NEC only used an infrared camera to acquire images. Although the investigators were able to derive thermal distributions for babies of various GA's, our previous study did not adopt a strict protocol for positioning the IR sensor in an optimal configuration, leading to more variations in image quality and size of region of interest. These limitations made it difficult to automate the selection of accurate regions of interest within the thermal image and required manual intervention from an image processing expert. In our current study the image acquisition process will be improved upon and standardized; a Microsoft Kinect sensor utilizing thermal, color and active depth vision will be integrated into a thermographic camera to allow for automated selection of regions of interest and will facilitate ease of use by medical personal at the patient's bedside. Embedded image processing algorithms will be developed to exploit the multi-spectral nature of collected information in order to automatically segment and unclutter the region of interest over which thermal distributions of relevance will be monitored.

Initial development of this system, including apparatus integration and assembly, rigorous camera calibration procedures in between the RGB-D and IR sensors, and complete testing will be performed in a laboratory environment, using only mock-up models of the clinical environment. Then, in order to fully validate the application of this automated infrared image processing system for early diagnosis of NEC, acquisition of relevant images from human subjects is needed. The study will be pursued in the NICU environment to further validate the newly developed non-invasive and automated imaging protocol on babies of various sizes and GA in the NICU, since NEC can occur in babies ranging from extremely premature to those born at term. Since the investigators hope to use this technology someday to help improve the diagnosis of NEC, the experimental imaging procedure will be performed on two patient populations. The investigators will first image the abdomens of a group of healthy newborns with the newly developed apparatus and image processing protocol to set up a standard thermal distribution. Then, the investigators will also image the abdomens of a group of newborns with proven NEC, using the same equipment. From the analysis of images from these two populations, the investigators will determine the suitability and necessary fine-tuning of the proposed automated infrared imaging tools, with the objective of augmenting the precision, usability, and reliability of the procedure for early diagnosis of NEC.

Objectives:

1. Develop an automated approach to the imaging and analysis of preterm infants' abdomens using thermographic sensors and embedded image processing algorithms.
2. Derive thermographic temperature distributions of the abdomens of two groups of babies: those who are normal (have no signs of feeding intolerance or signs of NEC), and those who have been diagnosed with proven NEC.
3. Determine if the automated image processing algorithms can discern a statistically significant difference between the normal thermal distributions and those derived from babies with NEC.