Executive Summary : | Radiotherapy is a crucial part of cancer treatments, but radiation-induced lung injury (RILI) is a severe complication and undesirable outcome. RILI includes radiation pneumonitis and radiation fibrosis, which cause acute lung tissue inflammation and chronic pulmonary tissue damage. Diagnosis is often made through clinical assessment and radiological findings, with the severity and extent of lung injuries influenced by patient, treatment, and tumor risk factors. Several researchers are developing machine learning (ML)-based models for predicting toxicity, but their application is hindered by low interpretability. This project proposes an AI-based Deep learning CNN architecture with auto encoder-decoder and global fusion modules for image reconstruction and classification. PET images are used for training, as PET can identify areas of damage that MRI or CT scans may not show. The system will be trained with both affected and unaffected lung images for identification and classification. The project aims to early predict lung damage due to radiation given for breast cancer on two levels using transfer learning. The first level uses pre-trained AI models to segment damaged lung regions and classify damage levels. The second level uses learning from the first level and other critical parameters to predict injury post-radiation in a new patient before the beginning of radiation. This system would help treat lungs during radiation, reducing the risk of injury if left untreated for the entire duration of radiation. The model will be validated by an oncologist and implemented in hardware for real-time testing. In the COVID era, this system would be beneficial to cancer patients, preventing further lung injury due to radiation. |
Co-PI: | Dr. Rajasekaran C, K. S. Rangasamy College Of Technology, Tamil Nadu-637215, Dr. Dhanalakshmi R, Indian Institute Of Information Technology Tiruchirappalli, Tamil Nadu-620012, Dr. SureshKumar Ramasamy, Fatima College Madurai, Tamil Nadu-625018 |