Predicting Response to Neoadjuvant Chemotherapy with PET Imaging Using Convolutional Neural Networks

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• 18F-FDG PET imaging is crucial for cancer diagnosis, staging, and monitoring treatment response.
• Predicting patient response to neoadjuvant chemotherapy before treatment is a significant clinical challenge.
• Radiomics, extracting quantitative features from medical images, offers a novel approach to tumor phenotyping.

Purpose of the Study:

• To investigate the potential of radiomics from pre-treatment 18F-FDG PET scans for predicting neoadjuvant chemotherapy response in esophageal cancer.
• To compare the performance of a convolutional neural network (CNN) approach against traditional radiomics methods using statistical classifiers.

Main Methods:

• A radiomics approach was employed, extracting numerous quantitative features from pre-therapy 18F-FDG PET scans.
• Two strategies were compared: statistical classifiers with over 100 imaging descriptors (including texture and SUV) and a 3S-CNN trained directly on PET image slices.
• The study included 107 patients with esophageal cancer.

Main Results:

• The 3S-CNN model demonstrated potential in extracting PET imaging features predictive of therapy response.
• The 3S-CNN achieved 80.7% sensitivity and 81.6% specificity in predicting non-responders.
• The CNN approach outperformed competing predictive models on the studied dataset.

Conclusions:

• Convolutional neural networks show promise for predicting neoadjuvant chemotherapy response using pre-treatment 18F-FDG PET scans in esophageal cancer.
• Automated feature learning with CNNs may offer a more effective radiomics strategy compared to traditional methods.
• This approach could lead to more personalized and effective cancer treatment strategies.