Development of Clinically High Efficient Platforms for Individualised Treatment of Cervix Cancer

Overview

Retrospective study utilizing patient data to develop and validate Machine Learning application. Available imaging data sets of patients who have completed treatment will be used to develop Normal tissue complication probability and Tumour control probability Hypothesis Integrating existing radiation treatment information, quantitative imaging and patient outcome data from completed and ongoing clinical trials will allow development of knowledge based systems for efficient treatment delivery and allow selection of patients for intensified treatment approaches in cervix cancer.

For Aim 1. Automatic delineation of complex tumour targets for cervical cancer for the Gross Tumour Volume (GTV) at baseline and at brachytherapy and High Risk Clinical Target Volume(CTV) at baseline and brachytherapy will be done on MRI. Following structures will be processed for automation on CT 1. Low Risk Clinical Target Volume (Low Risk CTV) 2. GTV: Nodal 3. Elective Nodal Pelvic Target Volume 4. Elective Nodal Pelvic and Paraaortic Volume 5. Rectum 6. Bladder 7. Sigmoid 8. Bowel 9. Bone Marrow For Aim 2. The Investigator intend to employ machine learning for developing more robust normal tissue toxicity prediction models. Further advanced techniques like texture analysis of radiation dose maps and follow up tissue density will also be performed to develop predictive models of toxicity. By using our patient datasets, Investigator want to create a library of proton beam plans with the proton planning systems that will be available in department of radiation oncology and using the developed normal tissue complication plots available the information of achievable doses through protons can help in identifying patients who will benefit from proton therapy. For Aim 3. Within this project Investigator intend to integrate staging, pathology and quantitative imaging texture features for response prediction and identification of "high risk cohort" in cervix cancer. Images and clinical data from patients that have MRI at baseline will be included The texture features can be used to categorise "good" and "poor responders" after chemoradiation. For the same cohort of patients the Investigator also have tissue available including results of additional biomarkers (like AKT,LICAM, PDL1,CD4 and CD8). The Investigator intend to first correlate difference in texture features and see if there is a pattern of different molecular features. In the second step imaging and molecular features could be integrated for developing" risk prediction models". GTV and HRCTV delineated on 150 data sets at baseline and brachytherapy within Aim 1 will be utilised to categorise responders and non-responders and validate another 150 patient data sets.