University of Twente Student Theses


Artificial intelligence based therapeutic response assessment of neoadjuvant immunotherapy for patients with bladder cancer

Greidanus, Joyce (2022) Artificial intelligence based therapeutic response assessment of neoadjuvant immunotherapy for patients with bladder cancer.

[img] PDF
Abstract:Introduction: Clinical studies showed promising results for neoadjuvant immunotherapy for patients with muscle invasive bladder cancer (MIBC). Unfortunately, only a subset of patients responds, urging the quest for predictive image features. We hypothesise that Artificial Intelligence (AI) can automatically quantify predictive image features for immunotherapy response. Patients and Method: A retrospective study is performed in which n=79 patients from the PURE-01 study and n=19 patients from the NABUCCO study were included. These patients were diagnosed with MIBC (staged ≥ cT2N0M0), and received neoadjuvant immunotherapy, followed by a radical cystectomy. For each patient, we analysed two Magnetic Resonance Imaging (MRI) scans; the first one was acquired before neoadjuvant immunotherapy (baseline) and the second one after therapy (on-treatment). Pathological treatment response was divided between pathological Complete Response (pCR) (ypT0N0Mx) and non-pCR. A nnU-Net model was trained to, automatically detect tumour. Volumetric analysis was performed to determine its predictive value for therapeutic response. Radiological tumour characteristics were extracted and a Random Forest Classifier was trained to identify predictive image features. Finally, a Convolutional Neural Network (CNN) was trained to classify pathological outcomes. Results: Segmentation volume analysis showed a higher predictive performance for the pathological outcome for the on-treatment volumes, compared to the baseline volumes (respectively 0.91 AUC and 0.81 AUC). The predictive value for the radiomic features of baseline, on-treatment and difference over time was 0.49 AUC, 0.70 AUC and 0.74 AUC, respectively. The CNN models overfit on the training data, the highest AUC score (0.62) showed no significant predictive performance. Conclusion: The results show that the predictive performance for image features, obtained after treatment, is promising for pathological response classification. These features might be used to indicate organ-sparring treatment after neoadjuvant therapy. Predictions based on tumour features, derived from the radiological images before neoadjuvant immunotherapy, remain challenging
Item Type:Essay (Master)
Faculty:TNW: Science and Technology
Subject:44 medicine, 50 technical science in general
Programme:Technical Medicine MSc (60033)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page