Project for a bioinformatics course taken at Sapienza. The task was to find a list of genes that are differentially expressed between case and control samples and characterise them. I used R to analyse a dataset of endometrium and endometriosis samples from GEO to identify differentially expressed genes which could be used for diagnosis and performed functional enrichment analysis on these genes to investigate the disease's pathogenesis.
Group project for a computer vision course taken at Sapienza. The task was to pre-process and extract features from the FaceForensics++ dataset, use these features to fine-tune a transformer model for DeepFake detection, and then investigate some model compression techniques. I worked on extracting depth estimates from the pre-processed video frames and fine-tuned some transformer models. Other group members worked on the pre-processing, extracting optical flows, and model compression using quantisation and distillation.
Assignment for a signal processing course taken at UoA. The task was to record time-series data during two different activities, extract relevant features from the data, and then create a decision model for a hypothetical health application. I chose to build an activity recognition model for classifying walking and running.
Thesis at UoA. I created scaled and subject-specific models of the lower limb of a typically developed child and a child with cerebral palsy to compare the estimated structure and function of the models using Python and OpenSim. I found that the differences in structure lead to differences in the estimated joint kinematics and muscle forces, highlighting the importance of using accurate geometries in musculoskeletal models. The methods established can also be used to investigate relationships between the accuracy of modelling methods and the severity of a child's cerebral palsy.
Project for an exercise science course taken at UoA. I investigated whether performing joint mobilisation tasks increased the risk of work-related musculoskeletal disorders (WMSDs) for physiotherapists. I analysed motion capture data using existing software to find RULA and REBA scores, low back compression forces, and joint strength capabilities at the most awkward posture for each task and subject. I then used R to perform two-way mixed ANOVA for each dependent variable. I found that all three tasks investigated can increase the risk of WMSDs for male and female physiotherapists.
Project for a computational techniques course taken at UoA. I wrote C++ functions to implement the power and deflation methods for finding eigenpairs / principal components of the coffee dataset. Next, I used Python to visualise the explained variance ratio, found that six principal components could represent the dataset, and reconstructed spectra for both coffee bean types. I projected the dataset onto the six most important principal components to find a pair that provided a good separation of the two coffee bean types.
Project for Summer Research Scholarship at the Auckland Bioengineering Institute. I used SOLIDWORKS to design a set of standard manipulation objects capable of encasing a Shimmer3 wireless multi-sensor unit and constructed two sets, one using Ecoflex 00-30 silicone and the other using 3D printed PLA plastic. I imported the Shimmer3 data into MATLAB to visualise the movement of the standard manipulation objects and was able to estimate object orientations in near real-time.