Jul 31, 2021
This research paper aimed to develop a deep learning Convolutional Neural Network (CNN) to predict the suitability of Deep Inspiration Breath Hold (DIBH) Intensity-Modulated Radiation Therapy (IMRT) for patients with Left-Sided Breast Cancer based on pre-treatment Chest X-Ray (CXR) alone.
The study utilized a dataset of Left-Sided Breast Cancer patients who underwent DIBH assessment after surgery and adjuvant chemotherapy. The assessment protocol involved three days of DIBH assessment followed by CT simulation, with specific acceptance criteria for DIBH suitability.
The primary objective was to label patients as suitable or not suitable for DIBH based on the assessment. The deep learning CNN model was trained on standard pre-RT CXR images acquired during inspiration breath-hold.
The methodology involved dividing the patients into training and testing cohorts, with image pre-processing and augmentation performed on the training cohort. The base 2D model architecture of VGG-19 was customized, and different optimizers and learning rates were explored to maximize model accuracy.
The results showed that the CNN architecture consisted of 16 2D layers in four groups, with relevant regions in the cardiac shadow, central mediastinum, and left upper lobe of the lung for DIBH prediction. The model achieved a mean accuracy of 0.93 ± 0.005 in the training set and an accuracy of 0.80 in the test set. The F1-score for predicting DIBH was 0.93 in the training set and 0.79 in the test set.
In conclusion, the deep learning method achieved an accuracy of 80% in predicting the suitability of DIBH based on CXR alone for patients with Left-Sided Breast Cancer. However, internal and external validation is necessary before clinical adoption
