PhD Defense: Amith J. Kamath Enhances Radiotherapy Treatment through Deep Learning

In this work, Amith addressed the limitation of traditional geometric metrics in anatomical contouring for radiotherapy. Radiation dosimetry—the measurement, calculation and assessment of the ionizing radiation dose absorbed by the human body—relies on high contouring quality. The traditional metrics correlate poorly with dose distribution and patient outcomes.

To address this issue, Amith advanced dosimetric contour quality assurance through three key efforts: (1) validating the clinical need for dosimetry-based quality assurance, (2) evaluating the reliability and efficiency of automated tools, and (3) developing novel applications to enhance clinical workflows.

In assessing the sensitivity of deep learning-based dose prediction models to segmentation variability, Amith found accurate dose reflection but variable performance across anatomical sites. In this context, he introduced two novel tools: ASTRA, a visualization platform for real-time dosimetric contour evaluation using sensitivity maps, and AutoDoseRank, a framework for automating segmentation prioritization based on clinical impact.

Congratulations to Dr. Amith J. Kamath on this incredible achievement and laying a foundation for intelligent quality assurance in next-generation radiotherapy treatment planning!

Publications:

Kamath, Amith; Willmann, Jonas; Andratschke, Nicolaus; Reyes, Mauricio (2025). The impact of U-Net architecture choices and skip connections on the robustness of segmentation across texture variations. Computers in Biology and Medicine 197(B):111056. Elsevier 10.1016/j.compbiomed.2025.111056

Willmann, Jonas; Kamath, Amith; Poel, Robert; Riggenbach, Elena; Mose, Lucas; Bertholet, Jenny; Muller, Silvan; Schmidhalter, Daniel; Andratschke, Nicolaus; Ermis, Ekin; Reyes, Mauricio (2025). Predicting the impact of target volume contouring variations on the organ at risk dose: results of a qualitative survey. Radiotherapy and Oncology 210:110999. Elsevier 10.1016/j.radonc.2025.110999

Poel, Robert; Kamath, Amith; Willmann, Jonas; Andratschke, Nicolaus; Ermis, Ekin; Aebersold, Daniel M.; Manser, Peter; Reyes, Mauricio (2023). Deep-Learning-Based Dose Predictor for Glioblastoma—Assessing the Sensitivity and Robustness for Dose Awareness in Contouring. Cancers 15(17):4226. MDPI AG 10.3390/cancers15174226

Kamath, Amith; Poel, Robert; Willmann, Jonas; Ermis, Ekin; Andratschke, Nicolaus; Reyes, Mauricio (2023). ASTRA: Atomic Surface Transformations for Radiotherapy Quality Assurance. Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC). IEEE 10.1109/EMBC40787.2023.10341062