Proof-of-principle approach for analysis of radiomic features in patients with ultracentral and central lung lesions treated using MR-guided hypofractionated radiotherapy

Committee/Faculty

Presenting Author(s)

MS

Maria L. Sandoval, MD MS

Radiation Oncology Resident, PGY-3
H. Lee Moffitt Cancer Center
Tampa, Florida, United States

Disclosure: I do not have any relevant financial / non-financial relationships with any proprietary interests.

Co-Author(s)

AS

Austin J. Sim, MD, JD

Department of Radiation Oncology
H. Lee Moffitt Cancer Center
Tampa, Florida, United States

Disclosure: ViewRay, Inc (Grant/Research Support) (Research Grant includes principal investigator, collaborator or consultant and pending grants as well as grants already received)
MT

Michal R. Tomaszewski, PhD

PhD
H. Lee Moffitt Cancer Center & Research Institute
Tampa, Florida, United States

Disclosure: Disclosure information not submitted.
JB

John M. Bryant, MD

Radiation Oncology Resident, PGY-2
H. Lee Moffitt Cancer Center
Tampa, Florida, United States

Disclosure: I do not have any relevant financial / non-financial relationships with any proprietary interests.
KL

Kujtmin Latifi, PhD

Department of Radiation Oncology
H. Lee Moffitt Cancer Center
Tampa, Florida, United States

Disclosure: Disclosure information not submitted.
RG

Robert Gillies, PhD

PhD
H. Lee Moffitt Cancer Center & Research Institute
Tampa, Florida, United States

Disclosure: Disclosure information not submitted.
EW

Evan J. Wuthrick, MD

Department of Radiation Oncology
H. Lee Moffit Cancer Center
Tampa, Florida, United States

Disclosure: Disclosure information not submitted.
Bradford A. Perez, MD

Department of Radiation Oncology
H. Lee Moffitt Cancer Center
Tampa, Florida, United States

Disclosure: AstraZeneca (Individual(s) Involved: Self): Consultant/Advisory Board; Bristol-Myers Squibb (Individual(s) Involved: Self): Consultant/Advisory Board, Research Grant; G1 Therapeutics (Individual(s) Involved: Self): Consultant/Advisory Board
TD

Thomas J. Dilling, MD

Department of Radiation Oncology
H. Lee Moffitt Cancer Center
Tampa, Florida, United States

Disclosure: Disclosure information not submitted.
SR

Stephen A. Rosenberg, MD MS

Department of Radiation Oncology
H. Lee Moffitt Cancer Center
Tampa, Florida, United States

Disclosure: Disclosure information not submitted.

Purpose: MR-guided radiotherapy (MRgRT) involves the acquisition of daily MR images for treatment, presenting an opportunity to develop MR-based biomarkers to predict response. We hypothesize that on treatment radiomic features in ultracentral and central lung lesions treated with MRgRT may correlate with clinical outcomes.

Methodology: We performed a single institution retrospective review of patients with ultracentral (per HILUS) or central (per RTOG) lung lesions treated with hypofractionated MRgRT. Images were acquired on a 0.35T MRI-linac system using a balanced steady state free precession pulse sequence at simulation (SIM) and prior to each treatment. Gross Tumor Volume (GTV) was segmented by the radiation oncologist on the SIM scan, and the daily pre-treatment scans were co-registered online to the SIM scan with attention to the GTV. MR images were normalized using the median ascending aorta signal intensity (blood pool). Image features (histogram and texture n=80, defined according to International Biomarker Standardization Initiative) were quantified in SIM, first (F1) and eight (F8) fraction images, and their ratios (F8/F1) quantified to capture imaging changes during treatment and evaluated for association with progression free survival (PFS) using univariate Cox proportional hazards model. All image analysis was performed using custom code in MATLAB 2019a.

Results: Between 3/2019 and 3/2021, 21 patients were treated with MRgRT with a median follow up of 11.7 months (range 1.6 – 24.4). Of these, 62% (n=13) had lung metastases and the remainder had primary lung cancers. In this cohort, 76% (n=16) had ultracentral lesions Group A per HILUS and all had central lesions per RTOG. Hypofractionated regimens included 60Gy in 8 fractions, 60Gy in 15 fractions or 50Gy in 10 fractions. The F8/F1 ratio identified 34 radiomic features with promising correlation to PFS in addition to features associated with PFS on the SIM scan. Table 1 demonstrates 8 representative features with lowest p-values. However, due to limited cohort size and preliminary nature of the analysis, no statistically significant relationships were observed following Bonferroni-Holm correction for multiple comparisons.

Table 1. Radiomic Features correlated with PFS
Radiomic Feature PFS SIM p-val PFS Rat8 p-val
Statistical Kurtosis 0.793 0.006*
Avg Coocurrence Difference variance 0.661 0.003**
Avg CoocurrenceContrast 0.460 0.007*
Avg 3D RLN (Run length non uniformity) 0.015* 0.843
GLSZM Large zone high grey level emphasis 0.008* 0.260
GLSZM Grey level non uniformity 0.017* 0.957
GLSZM Zone size non uniformity 0.014* 0.278
NGTDM Busyness 0.038* 0.004**
* p-val < 0.05, ** p-val < 0.005

Conclusions: This proof of principle study revealed good correlation of several radiomic texture features to PFS. We are currently validating these preliminary findings in a larger cohort. In the era of personalized medicine with MRgRT, radiomics may allow early identification of treatment response and allow treatment intensity customization to improve outcomes.

References: