Background

It is useful to predict planned dosimetry and determine the eligibility of a liver cancer patient for SBRT treatment using knowledge based planning (KBP). We compare the predictive accuracy using the overlap volume histogram (OVH) and statistical voxel dose learning (SVDL) KBP prediction models for coplanar VMAT to non-coplanar 4π radiotherapy plans.

Methods

In this study, 21 liver SBRT cases were selected, which were initially treated using coplanar VMAT plans. They were then re-planned using 4π IMRT plans with 20 inversely optimized non-coplanar beams. OVH was calculated by expanding the planning target volume (PTV) and then plotting the percent overlap volume v with the liver vs. r _v , the expansion distance. SVDL calculated the distance to the PTV for all liver voxels and bins the voxels of the same distance. Their dose information is approximated by either taking the median or using a skew-normal or non-parametric fit, which was then applied to voxels of unknown dose for each patient in a leave-one-out test. The liver volume receiving less than 15 Gy (V_<15Gy), DVHs, and 3D dose distributions were predicted and compared between the prediction models and planning methods.

Results

On average, V_<15Gy was predicted within 5%. SVDL was more accurate than OVH and able to predict DVH and 3D dose distributions. Median SVDL yielded predictive errors similar or lower than the fitting methods and is more computationally efficient. Prediction of the 4π dose was more accurate compared to VMAT for all prediction methods, with significant (p < 0.05) results except for OVH predicting liver V_<15Gy (p = 0.063).

Conclusions

In addition to evaluating plan quality, KBP is useful to automatically determine the patient eligibility for liver SBRT and quantify the dosimetric gains from non-coplanar 4π plans. The two here analyzed dose prediction methods performed more accurately for the 4π plans than VMAT.

Purpose

The 4π static non-coplanar radiotherapy delivery technique has demonstrated better normal tissue sparing and dose conformity than the clinically used volumetric modulated arc therapy (VMAT). It is unclear whether this is a fundamental limitation of VMAT delivery or the coplanar nature of its typical clinical plans. The dosimetry and the limits of normal tissue toxicity constrained dose escalation of coplanar VMAT, non-coplanar VMAT and 4π radiotherapy are quantified in this study.

Methods and materials

Clinical stereotactic body radiation therapy plans for 20 liver patients receiving 30-60 Gy using coplanar VMAT (cVMAT) were re-planned using 3-4 partial non-coplanar arcs (nVMAT) and 4π with 20 intensity-modulated non-coplanar fields. The conformity number (CN), homogeneity index (HI), 50% dose spillage volume (R₅₀), normal liver volume receiving >15 Gy (VL_>15), dose to organs at risk (OARs), and tumor control probability (TCP) were compared for all three treatment plans. The maximum tolerable dose (MTD) yielding a normal liver normal tissue control probability (NTCP) below 1%, 5%, and 10% was calculated with the Lyman-Kutcher-Burman model for each plan, as well as the resulting survival fractions at one, two, three, and four years.

Results

Compared to cVMAT, the nVMAT and 4π plans reduced VL_>15 by an average of 5 cm³ and 80 cm³, respectively. 4π reduced the 50% dose spillage volume by ~23% compared to both VMAT plans, and either significantly decreased or maintained OAR doses. The 4π MTDs and survival fractions were significantly higher than both cVMAT and nVMAT (p<0.05) for all normal liver NTCP limits used in this study.

Conclusions

The 4π technique provides significantly better OAR sparing than both cVMAT and vMAT and enables more clinically relevant dose escalation for tumor local control. Therefore, despite the current accessibility of nVMAT, it is not a viable alternative to 4π for liver SBRT.

Purpose

To predict the neoadjuvant chemoradiation therapy (CRT) response in patients with locally advanced rectal cancer (LARC) using radiomics and deep learning based on pre-treatment MRI and a mid-radiation follow-up MRI taken 3-4 weeks after the start of CRT.

Methods

A total of 51 patients were included, 45 with pre-treatment, 41 with mid-radiation therapy (RT), and 35 with both MRI sets. The multi-parametric MRI protocol included T2, diffusion weighted imaging (DWI) with b-values of 0 and 800 s/mm², and dynamic-contrast-enhanced (DCE) MRI. After completing CRT and surgery, the specimen was examined to determine the pathological response based on the tumor regression grade. The tumor ROI was manually drawn on the post-contrast image and mapped to other sequences. The total tumor volume and mean apparent diffusion coefficient (ADC) were measured. Radiomics using GLCM texture and histogram parameters, and deep learning using a convolutional neural network (CNN), were performed to differentiate pathologic complete response (pCR) vs. non-pCR, and good response (GR) vs. non-GR.

Results

Tumor volume decreased and ADC increased significantly in the mid-RT MRI compared to the pre-treatment MRI. For predicting pCR vs. non-pCR, combining ROI and radiomics features achieved an AUC of 0.80 for pre-treatment, 0.82 for mid-RT, and 0.86 for both MRI together. For predicting GR vs. non-GR, the AUC was 0.91 for pre-treatment, 0.92 for mid-RT, and 0.93 for both MRI together. In deep learning using CNN, combining pre-treatment and mid-RT MRI achieved a higher accuracy compared to using either dataset alone, with AUC of 0.83 for predicting pCR vs. non-pCR.

Conclusion

Radiomics based on pre-treatment and early follow-up multi-parametric MRI in LARC patients receiving CRT could extract comprehensive quantitative information to predict final pathologic response.