Evaluation of Neoadjuvant Chemoradiotherapy Response in Rectal Cancer Using MR Images and Deep Learning Neural Networks.

INTRODUCTION: The aim of the study was to develop deep-learning neural networks to guide treatment decisions and for the accurate evaluation of tumor response to neoadjuvant chemoradiotherapy (nCRT) in rectal cancer using magnetic resonance (MR) images. METHODS: Fifty-nine tumors with stage 2 or 3 rectal cancer that received nCRT were retrospectively evaluated. Pathological tumor regression grading was carried out using the Dworak (Dw-TRG) guidelines and served as the ground truth for response predictions. Imaging-based tumor regression grading was performed according to the MERCURY group guidelines from pre-treatment and post-treatment para-axial T2-weighted MR images (MR-TRG). Tumor signal intensity signatures were extracted by segmenting the tumors volumetrically on the images. Normalized histograms of the signatures were used as input to a deep neural network (DNN) housing long short-term memory (LSTM) units. The output of the network was the tumor regression grading prediction, DNN-TRG. RESULTS: In predicting complete or good response, DNN-TRG demonstrated modest agreement with Dw-TRG (Cohen's kappa= 0.79) and achieved 84.6% sensitivity, 93.9% specificity, and 89.8% accuracy. MR-TRG revealed 46.2% sensitivity, 100% specificity, and 76.3% accuracy. In predicting a complete response, DNN-TRG showed slight agreement with Dw-TRG (Cohen's kappa= 0.75) with 71.4% sensitivity, 97.8% specificity, and 91.5% accuracy. MR-TRG provided 42.9% sensitivity, 100% specificity, and 86.4% accuracy. DNN-TRG benefited from higher sensitivity but lower specificity, leading to higher accuracy than MR-TRG in predicting tumor response. CONCLUSION: The use of deep LSTM neural networks is a promising approach for evaluating the tumor response to nCRT in rectal cancer.

Evaluating complete response prediction rates in locally advanced rectal cancer with different radiomics segmentation approaches.

Purpose: Studies examining prediction of complete response (CR) in locally advanced rectum cancer (LARC) from pre/post chemoradiotherapy (CRT) magnetic resonance imaging (MRI) are performed mostly with segmentations of the tumor, whereas only in two studies segmentation included tumor and mesorectum. Additionally, pelvic extramesorectal region, which is included in the clinical target volume (CTV) of radiotherapy, may contain information. Therefore, we aimed to compare predictive rates of radiomics analysis with features extracted from segmentations of tumor, tumor+mesorectum, and CTV. Methods and materials: Ninety-three LARC patients who underwent CRT in our institution between 2012 and 2019 were retrospectively scanned. Patients were divided into CR and non-CR groups. Tumor, tumor+mesorectum and CTV were segmented on T2 preCRT MRI images. Extracted features were compared for best area under the curve (AUC) of CR prediction with 15 machine-learning models. Results: CR was observed in 25 patients (26.8%), of whom 13 had pathological, and 12 had clinical complete response. For tumor, tumor+mesorectum and CTV segmentations, the best AUC were 0.84, 0.81, 0.77 in the training set and 0.85, 0.83 and 0.72 in the test set, respectively; sensitivity and specificity for the test set were 76%, 90%, 76% and 71%, 67% and 62%, respectively. Conclusion: Although the highest AUC result is obtained from the tumor segmentation, the highest accuracy and sensitivity are detected with tumor+mesorectum segmentation and these findings align with previous studies, suggesting that the mesorectum contains valuable insights for CR. The lowest result is obtained with CTV segmentation. More studies with mesorectum and pelvic nodal regions included in segmentation are needed.

Investigation of approaches for internal target volume definition using 4-dimensional computed tomography in stereotactic body radiotherapy of lung cancer.

The present study was undertaken to investigate the suitability of alternative internal target volume (ITV) delineation strategies based on maximum intensity projection (MIP), average intensity projection (AIP), 2 extreme phases and 4 phases images relative to the ITV10phase in stereotactic body radiation therapy (SBRT) for lung cancer. The 4-dimensional computed tomography (4DCT) data of 15 lung cancer patients treated with SBRT in our clinic were used. Five different ITVs were generated as follows: merging GTVs from 10 phases (ITV10Phase); merging GTVs from 2 extreme phases (0%, 50%) (ITV2Phase); merging GTVs from 4 phases (0%, 20%, 50%, and 70%) (ITV4Phase); delineating GTV on MIP (ITVMIP), and delineating GTV on AIP (ITVAIP). PTV10Phase, PTV2Phase, PTV4Phase, PTVMIP, and PTVAIP were generated by adding a 5-mm margin around the related ITV. Volumetric analyses were performed for 4 ITVs and PTVs relative to ITV10phase and PTV10phase. SBRT plans made for all PTVs were evaluated for dosimetric effect of alternative ITV delineation strategies. The mean percentage overlap volume (POV) for PTV2phase, PTV4phase, PTVMIP, and PTVAIP relative to PTV10phase were 84.2 ± 5.4%, 92.0 ± 2.9%, 82.2 ± 5.7%, and 73.8 ± 9.3%, for lower-lobe tumors, respectively. The mean POV for PTV2phase, PTV4phase, PTVMIP, and PTVAIP relative to PTV10phase were 93.2 ± 2.5%, 95.9 ± 1.0%, 87.5 ± 6.7%, and 83.3 ± 6.8% for upper-lobe, respectively. For lower-lobe tumors the mean differences in V20 and MLD for plans based on PTV2phase and PTV4phase were <0.5% and <10 cGy, compared with a plan based on PTV10phase. The use of PTV based on 4 respiratory phases and a 5-mm margin is a safe approach to reduce the workload of target delineation for tumors located in both lower and upper lobes.