Click here to

Session: Science Council Session: Innovative Technologies to Advance Diagnosis and Treatment [Return to Session]

An Imaging Based Biological Framework for Personalizing Radiation Therapy of Lung Cancer Using Interpretable Bayesian Networks and NTCP-Based Treatment Planning

A Ajdari1*, T Bortfeld2, (1) Harvard Medical School & Massachusetts General Hospital, Boston, MA, (2) Massachusetts General Hospital, Boston, MA

Presentations

TU-EF-TRACK 4-1 (Tuesday, 7/27/2021) 3:30 PM - 5:30 PM [Eastern Time (GMT-4)]

Purpose: To develop a biological framework for adapting radiation therapy (RT) treatment plans of lung cancer patients by integrating functional imaging, machine learning, and convex optimization.

Methods: Seventy-four non-small cell lung cancer (NSCLC) patients treated with passive scattered proton (PSP, n=28) and intensity modulated RT (IMRT, n=46) were retrospectively analyzed. Baseline and mid-treatment [18]F-fluorodeoxyglucose (FDG) positron emission tomography (PET) imaging information was integrated with clinicopathological/dosimetric information to estimate the patient-specific response. A novel Bayesian Network (BN) was trained and validated for predicting grade2+ radiation pneumonitis (RP) at three months post-RT. BN predictions were used to estimate an individualized normal tissue toxicity probability (NTCP) function for each patient. The updated NTCP functions were then integrated in the RT inverse optimization and a convex optimization algorithm was developed to solve the resulting problem. The framework was tested on 20 patients (PSP, n=10; IMRT, n=10) to optimally adapt the dose distribution according to mid-treatment biological information.

Results: BN model was able to predict the RP risk with an AUC of 0.82 (95% CI=[0.72-0.91]). The treatment plans for 17 patients (85%) were adapted to keep the updated RP risk below 20% (9 in PST and 8 in IMRT groups). Treatment plans for patients whose predicted RP risk exceed the predefined threshold were de-escalated (n=14, average target dose reduction = 5.4Gy). Patients with predicted NTCP lower than 20% saw an increase in average tumor dose after RT adaptation (n=6, average target dose increase = 4.1Gy). Overall, 17% decrease in predicted RP risk was achieved using the proposed methodology (SD=7.4%). No difference between PSP and IMRT groups were observed in terms of RP risk prediction or plan adaptation.

Conclusion: Mid-treatment adaptation could significantly lower the risk of RP without compromising tumor control. The proposed methodology could help the design of personalized clinical trials for NSCLC patients.

Handouts

    Keywords

    FDG PET, Optimization, NTCP

    Taxonomy

    Not Applicable / None Entered.

    Contact Email

    Share: