Soodeh Ahani

Experience

• 

  NeuroWave Systems Inc.

  Apr 2016 - Jul 2018

  Data Science Algorithm Engineer/ Postdoctoral researcher @ UBC

  • Burst and suppression (BS) pattern detection in electroencephalogram (EEG) for automated depth of anesthesia control• Pioneered innovative features utilizing the smoothed L0 norm for Burst Suppression (BS) pattern classification in EEG signals, concurrently creating an AdaBoost machine-learning model for automated BS detection in surgical settings• Attained a clinically significant 2.55% enhancement in F1-score (89.77% for our proposed method) compared to NeuroSENSE monitors (87.22%), demonstrating impactful advancements (i.e., 22 minutes fewer false alarms over 45.86 hours of test data)• Implemented and validated this improvement across 192 hours of EEG data from 90 vulnerable elderly patients under general anesthesia, bolstering reliability and safety in drug titration with the potential for substantial improvements in patient outcomes• Designed and implemented an intuitive GUI-based labeling prototype, optimizing the efficiency of manual annotation through a user-friendly visual interface. Show less

• 

  Medtronic

  Jun 2018 - now

  MITACS ELEVATE - Postdoctoral fellow @UBC

  • Non-contact vital sign detection and video monitoring for infants in the NICU, including data collection and analysis• Initiated and led the development of the study, overseeing graduate students, clinical staff, and stakeholders to ensure project success and timely delivery of objectives• Developed and deployed transfer learning-based object detection models (YOLOv5 and YOLOv7Face) for face recognition and tracking• Engineered algorithms for vital sign estimation, employing advanced signal processing, and deep learning methods• Enhanced intraclass correlation coefficient to 0.91 (95% CI: 0.89 − 0.93), achieving 14% higher reliability than the reference ICC between nurses and neonatologists (i.e., ICC= 0.77)• Substantially improved Bland-Altman analysis, narrowing the limit of agreement (LOA) to ±14.36, surpassing LOA values of impedance sensors (±20.96) and a former EVM-based method (±22.8) Show less

• 

  The University of British Columbia, Medtronic, and Novarc Technologies Inc.

  Feb 2019 - Jul 2023

  Data Management Experience @ UBC

  • Managed a team of engineers, clinical staff, and our data collection team at our industrial collaborator site to run both our health care and industrial welding projects simultaneously• Managed end-to-end data processes for both projects, including ethics application drafting for our vital sign detection project at BC Women's Hospital NICU to data collection, quality assessment, cleaning, storage, analysis, and visualization for both projects, encompassing video, audio, and structured data from Case Report Forms (CRFs)• Demonstrated senior-level leadership in fostering collaborative and motivating work environments for diverse stakeholders Show less

• 

  The University of British Columbia

  Feb 2022 - now

  Data Science Researcher

  1) Non-contact vital sign detection and video monitoring for infants in the NICU, including data collection and analysis.Industrial partner: Medtronic.• Non-contact vital sign detection and video monitoring for infants in the NICU, including data collection and analysis• Initiated and led the development of the study, overseeing graduate students, clinical staff, and stakeholders to ensure project success and timely delivery of objectives• Developed and deployed transfer learning-based object detection models (YOLOv5 and YOLOv7Face) for face recognition and tracking• Engineered algorithms for vital sign estimation, employing advanced signal processing, and deep learning methods• Enhanced intraclass correlation coefficient to 0.91 (95% CI: 0.89 − 0.93), achieving 14% higher reliability than the reference ICC between nurses and neonatologists (i.e., ICC= 0.77)• Substantially improved Bland-Altman analysis, narrowing the limit of agreement (LOA) to ±14.36, surpassing LOA values of impedance sensors (±20.96) and a former EVM-based method (±22.8)2) Welding robot quality assessment using acoustic and vision data.Industrial partner: Novarc Technologies Inc.• Welding robot quality assessment using acoustic and vision data• Utilized my advanced skills in data management and senior-level leadership to lead the design of the study and develop state-of-the-art deep learning models (CNN, ResNet 18 2D, and 3D) for innovative defect detection and seam tracking, leveraging both visual and audio data• Delivered exceptional results in defect detection and robotic welding quality improvement, achieving a stellar F1 score (LOP 99%, LOF 99%, Undercut 99%, Cold Lap 99%, LOP 99%)• Developed recurrent neural networks (RNN) models to analyze welding current and voltage time-series data Show less