Treatment Response Prediction for Colorectal Liver Metastases Using Pre-treatment Computed Tomography Scans
Loading...
Authors
Piliposyan, Mane
Date
2025-01-31
Type
thesis
Language
eng
Keyword
Colorectal Cancer Liver Metastasis , Deep Learning , Radiomics , Sel-Supervised Learning , Treatment Response , Prediction , Transfer Learning
Alternative Title
Abstract
Colorectal cancer is the third most common cancer globally, with a high mortality rate due to metastatic progression, particularly in the liver. Early diagnosis and effective treatment are critical to improving survival rates. Surgical resection remains the cornerstone of curative therapy, but only a small subset of patients are eligible for surgery due to the already advanced stage of the disease at the time of diagnosis. For patients with initially unresectable liver metastases, neoadjuvant chemotherapy can downstage tumours, potentially making surgical resection feasible. This approach reduces tumour size, which helps to lower the risk of recurrence. Predicting treatment response is crucial to optimizing treatment strategies, as it can guide personalized approaches and avoid exposing patients to the toxicity of chemotherapy.
In this study, we explored two approaches to developing predictive models for predicting treatment response in patients with initially unresectable colorectal liver metastases (CRLM). The models used single pre-treatment contrast-enhanced computed tomography scans with response categories determined based on volumetric thresholds. The first approach utilized traditional radiomics to extract quantitative image features, which were then analyzed using machine learning algorithms to predict treatment outcomes. The second approach employed Deep Learning (DL) techniques, specifically convolutional neural networks and transfer learning, to automatically learn and extract image-based features directly from the scans. Both approaches provide non-invasive tools to assist clinical decision-making and tailor personalized treatment strategies for CRLM patients. The study included 355 patients with initially unresectable CRLM.
The main contribution of the study is the development of predictive models for treatment outcomes before the initiation of chemotherapy. We demonstrate that baseline computed tomography scans contain valuable information that enables the assessment of early prognosis for treatment response. The radiomic-based methods performed significantly better by quantifying tumour heterogeneity and achieved an area under the curve of 0.77. While the DL model was not as successful, it was still able to capture a moderate signal with an area under the curve values ranging around 0.60.