Graduate Students Seminar
|Session Chair:||Mingkai Yu|
|Discussant:||Dr. Osman Guler|
Speaker 1: Carlos Barajas
- Neural Networks for Sanitization of Compton Camera Based Prompt Gamma Imaging Data for Proton Radiotherapy
- Real-time imaging has potential to greatly increase the effectiveness of proton beam therapy for cancer treatment. One promising method of real-time imaging is the use of a Compton camera to detect prompt gamma rays, which are emitted along the path of the beam, in order to reconstruct their origin. However, due to limitations in the Compton camera's ability to detect prompt gamma interactions, the data are often ambiguous, making reconstructions based on them unusable for practical purposes. A neural network's ability to detect subtleties in data that traditional models do not use make it one possible candidate for the improvement of classification of Compton camera data. Preliminary work shows how challenging the problem is with an accuracy of only 70. Improvements in data pre-processing and the use of progressively more sophisticated and complex neural networks gives us the chance to explore both fully connected and recurrent neural networks.
Speaker 2: Rabab Elnaiem
- Some criteria and tests for the assessment of bioequivalence and biosimilarity
- Bioequivalence testing deals with assessing the similarity of two drug products with respect to the rate and extent to which the active drug ingredient is absorbed into the blood, and becomes available at the site of drug action.
- A common bioequivalence criterion is that of average bioequivalence (ABE), which deals with testing if the difference between appropriate mean responses is within predefined equivalence limits. The response that is typically used is the ln(AUC); the natural logarithm of the area under the time-concentration curve. A popular test for ABE is the two one-sided t-test (TOST), introduced by Schuirmann (1981). However, the TOST is very conservative when the variability becomes large. The conservatism of the TOST can be easily fixed by applying a bootstrap calibration. The type I error then becomes close to the nominal level, giving significant gain in power. Consequently, the bootstrap-calibrated TOST results in a reduction in sample size. A scaled criterion is also considered for testing the bioequivalence of highly variable drug products.
- For assessing bioequivalence, instead of focusing on a criterion based on averages, an evaluation of the similarity between the distributions of the responses can be used as an alternative criterion. We propose the use of the overlap coefficient (OVL), which represents the area of overlap between two probability distributions, as a measure of the similarity between distributions. Using a fiducial approach, we have explored the computation of confidence limits for the OVL value. The confidence limits can be used to decide if the OVL value is large enough in order to declare similarity.