| 1. |
- Zhang, Tianfang
(författare)
-
Probabilistic machine learning methods for automated radiation therapy treatment planning
- 2021
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- In this thesis, different parts of an automated process for radiation therapy treatment planning are investigated from a mathematical and computational perspective. Whereas traditional inverse planning is labor-intensive, often comprising several reiterations between treatment planner and physician before a plan can be approved, much of recent research have been aimed at using a data-driven approach by learning from historically delivered plans. Such an automated planning pipeline is commonly divided into a first part of predicting achievable values of dose-related quantities, and a second part of finding instructions to the treatment machine mimicking as best as possible the predicted values. Challenges associated with this type of prediction–mimicking workflow exist, however—for example, in typical applications, patient data is high-dimensional, scarce and has relatively low signal-to-noise ratio due to inter-planner variations, and significant information may be lost in the transition between prediction and mimicking.We propose to address these challenges through better probabilistic modeling of the predictive inferences of dose-related quantities and increased accuracy of the optimization functions used for dose mimicking. In particular, starting with the disconnect between conventional planning objectives and evaluation metrics, in the first paper, we establish a framework for handling dose statistics as optimization function constituents. Subsequently, in the second and fourth papers, we present ways of predicting spatial dose and dose statistics, respectively, in a probabilistically rigorous fashion, the latter application relying on the similarity-based mixture-of-experts model developed in the third paper. As a nonparametric Bayesian regression model, equipped with a mean-field and stochastic variational inference algorithm, this mixture-of-experts model is suitable for managing complex input–output relationships and skewed or multimodal distributions. The second and fourth papers also introduce dose mimicking objectives able to leverage predictive distributions of spatial dose and dose statistics. In the fifth paper, we further build upon the probabilistic paradigm, merging the fields of multicriteria optimization and automated planning to create a semiautomatic alternative workflow in which certain manual intervention is possible. Lastly, in the sixth paper, we present a means of incorporating robustness against geometric uncertainties into an automated planning pipeline.
|
|
| 2. |
- Westerberg, Marcus, 1990-
(författare)
-
Prostate cancer incidence, treatment and mortality : Empirical longitudinal register-based studies and methods for handling missing data
- 2022
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- The diagnostic activity for prostate cancer has increased substantially in Sweden, primarily due to increased use of prostate-specific antigen (PSA) testing in asymptomatic men, and this has led to a large increase in diagnoses. There have also been changes in the diagnostic workup, guidelines, treatment strategies, and more effective treatments have been introduced in different phases of the disease. This thesis aims to increase the understanding of consequences of changes in diagnostic activity and treatment, with a focus on empirical studies, methodological development, and handling of missing data.In paper I, the survival of men with metastatic prostate cancer was investigated across calendar time periods by use of Kaplan-Meier analyses and Cox regression. The median survival from diagnosis increased with six months comparing men diagnosed 1998-2001 with men diagnosed 2010-2015, while median PSA decreased.In paper II, a discrete time multivariate longitudinal model was combined with a proxy for the unobserved level of diagnostic activity to produce prognoses of incidence and mortality. Simulations indicated that a higher diagnostic activity was associated with fewer men diagnosed with metastatic disease and fewer prostate cancer deaths.In paper III, we looked for clinical variables predictive of the survival of men with castration-resistant prostate cancer (CRPC). A new data base was created including longitudinal data on prescriptions of hormonal treatment, PSA, and cause of death. We found that PSA doubling time and PSA at time of CRCP were highly predictive and could be used for treatment decision.In paper IV, we estimated annual incidence of metastatic prostate cancer using different methods for handling missing data in metastatic status (M stage). Missing data in M stage was high and varied over calendar time and risk groups, yet each method indicated a downward trend in incidence. Although men with unknown metastatic status cannot be assumed to have nonmetastatic disease in general, this may be reasonable among those with tumour characteristics that indicate a low risk of metastases.In paper V, the estimation of multivariate longitudinal models was considered in a context where some events are observed on a coarser level (e.g. grouped) at some time points, causing gaps in the data. The likelihood function, score and observed information were derived under an independent coarsening mechanism. A simulation study was conducted comparing properties of several estimators including direct maximum likelihood and Monte Carlo Expectation Maximisation.
|
|
