Samanah Rabienia Haratbar, a doctoral candidate in biomedical engineering, took third place for her poster titled, "A Machine Learning Approach for the Prediction of Retinopathy of Prematurity (ROP) in Preterm Infants"
Retinopathy of prematurity (ROP) is a leading cause of visual impairment in preterm infants. Predicting ROP plays a vital role in preventing vision loss. The objective of this study is to employ a machine learning algorithm with influencing factors (e.g., gestational age, birth weight, small for gestational age) as inputs to predict ROP. Methods: Data were collected from 230 preterm infants (23 0/7 to 34 6/7 weeks gestation) at the Kentucky Children's Hospital, including 200 infants without ROP and 30 infants with ROP. A logistic regression-based model for predictive analysis was used to predict ROP. Model training was performed using seven independent variables including gestational age, birth weight, small for gestational age, gender, prenatal steroids, cesarean section, and multiple gestation. All analyses were performed using Python program and a data analysis tool of Pandas. The model performance was examined using a metrics including the sensitivity, specificity, area under the receiver operating characteristic curve (ROC), and harmonic mean of the model’s precision (F-score). Results: Our logistic regression model predicts the ROP with the sensitivity of 0.74, specificity of 0.83, area under ROC of 0.86, and F-score of 0.52. Among seven independent variables, gestational age is the most significant factor for ROP prediction, which meets the clinical expectation. Conclusions: With the promising logistic regression model established in this pilot study, we are now adding other influencing factors such as intermittent hypoxemia for better prediction and management of ROP.
Mehrana Mohtasebi, a doctoral candidate in biomedical engineering, received the Judges Choice Award for her poster titled, "Noncontact Optical Assessment of Disrupted Cerebral Functional Connectivity in a Piglet Model of Transient Ischemic Stroke"
Perinatal ischemic stroke results from the lack of blood supply to brain tissue, possibly leading to cerebral ischemic/hypoxic stress, neurological disorder, and brain network impairment. Preterm infants with ischemic stroke are prone to alterations in cerebral blood flow (CBF) and associated spontaneous low-frequency oscillations (LFOs). However, there are no established noninvasive imaging methods for continuous monitoring of CBF alterations at the bedside in neonatal intensive care units (NICUs). An innovative camera-based speckle contrast diffuse correlation tomography (scDCT) technology has been recently developed in our laboratory, which enables noncontact, noninvasive, and high-density 3D imaging of CBF distributions in cerebral cortex. In the present study, the capability of scDCT technique for 3D imaging of CBF distributions in a neonatal piglet model of transient ischemic stroke was demonstrated. Moreover, power spectral density analyses of LFOs and the network connections over the brain were assessed before and after the induction of acute ischemic stroke. The stroke resulted in a substantial decrease in CBF, attenuations in resting-state LFOs over the LFO frequency band (0.01–0.08 Hz), and functional connectivity disruptions in motor and somatosensory cortices. This pilot study demonstrated the feasibility and safety of scDCT for noninvasive detection of resting-state LFO alteration and functional connectivity disruption after stroke. We are currently testing this fully noncontact scDCT technology for 3D imaging of brain hemodynamics in the NICU, with the ultimate goal of instantly evaluating and managing brain injury/health to improve the clinical decision and outcome.
The awards were generously sponsored by the Institute for the Study of Free Enterprise in the Gatton College of Business & Economics.