Integrating Deep Learning-Based Dose Distribution Prediction with Bayesian Networks for Decision Support in Radiotherapy for Upper Gastrointestinal cancer.

Purpose: Selecting the better techniques to harbor optimal motion management, either a stereotactic linear accelerator delivery using TrueBeam (TBX) or Magnetic Resonance (MR)-guided gated delivery using MRIdian (MRG), is time-consuming and costly. To address this challenge, we aimed to develop a decision-supporting algorithm based on a combination of deep learning-generated dose distributions and clinical data. Materials and Methods: We retrospectively analyzed 65 patients with liver or pancreatic cancer who underwent both TBX and MRG simulations and planning process. We trained three-dimensional U-Net deep learning models to predict dose distributions and generated dose volume histograms (DVHs) for each system. We integrated predicted DVH metrics into a Bayesian network (BN) model incorporating clinical data. Results: The MRG prediction model outperformed the TBX model, demonstrating statistically significant superiorities in predicting normalized dose to the PTV and liver. We developed a final BN prediction model integrating the predictive DVH metrics with patient factors like age, PTV size, and tumor location. This BN model an area under the receiver operating characteristic curve index of 83.56%. The decision tree derived from the BN model showed that the tumor location (abutting vs. apart of PTV to hollow viscus organs) was the most important factor to determine TBX or MRG. Conclusion: We demonstrated a decision-supporting algorithm for selecting optimal RT plans in upper gastrointestinal cancers, incorporating both deep learning-based dose prediction and BN-based treatment selection. This approach might streamline the decision-making process, saving resources and improving treatment outcomes for patients undergoing RT.

Locoregional Recurrence in Adenoid Cystic Carcinoma of the Breast: A Retrospective, Multicenter Study (KROG 22-14).

Purpose: This study aims to evaluate the treatment approaches and locoregional patterns for Adenoid cystic carcinoma (ACC) in the breast, which is an uncommon malignant tumor with limited clinical data. Materials and Methods: A total of 93 patients diagnosed with primary ACC in the breast between 1992 and 2022 were collected from multi-institutions. All patients underwent surgical resection, including breast-conserving surgery (BCS) or total mastectomy (TM). The recurrence patterns and locoregional recurrence-free survival (LRFS) were assessed. Results: Seventy-five patients (80.7%) underwent BCS, and 71 of them (94.7%) received post-operative radiation therapy (PORT). Eighteen patients (19.3%) underwent TM, with 5 of them (27.8%) also receiving PORT. With a median follow-up of 50 months, the LRFS rate was 84.2% at 5 years. Local recurrence (LR) was observed in 5 patients (5.4%) and 4 cases (80%) of the LR occurred in the tumor bed. Three of LR (3/75, 4.0%) had a history of BCS and PORT, meanwhile, two of LR (2/18, 11.1%) had a history of mastectomy. Regional recurrence occurred in 2 patients (2.2%), and both cases had a history of PORT with (n=1) and without (n=1) irradiation of the regional lymph nodes. Partial breast irradiation (p=0.35), BCS (p=0.96) and PORT in BCS group (p=0.33) had no significant association with LRFS. Conclusion: BCS followed by PORT was the predominant treatment approach for ACC of the breast and local recurrence mostly occurred in the tumor bed. The findings of this study suggest that partial breast irradiation might be considered for PORT in primary breast ACC.

Evaluation of cosmetic outcomes in breast reconstruction patients undergoing radiotherapy using an anomaly generative adversarial network model.

Considering the rising prevalence of breast reconstruction followed by radiotherapy (RT), evaluating the cosmetic impact of RT is crucial. Currently, there are limited tools for objectively assessing cosmetic outcomes in patients who have undergone reconstruction. Therefore, we validated the cosmetic outcome using a previously developed anomaly Generative Adversarial Network (GAN)-based model and evaluated its utility. Between January 2016 and December 2020, we collected computed tomography (CT) images from 82 breast cancer patients who underwent immediate reconstruction surgery followed by radiotherapy. Among these patients, 38 received immediate implant insertion, while 44 underwent autologous breast reconstruction. Anomaly scores (AS) were estimated using an anomaly GAN model at pre-RT, 1st follow-up, 1-year (Post-1Y) and 2-year (Post-2Y) after RT. Subsequently, the scores were analyzed in a time-series manner, considering reconstruction types (implant versus autologous), RT techniques, and the incidence of major complications. The median age of the patients was 46 years (range 29-62). The AS between Post-1Y and Post-2Y demonstrated a positive relationship (coefficient 0.515, P < 0.001). The AS was significantly associated with objective cosmetic indices, namely Breast Contour Difference (P = 0.009) and Breast Area Difference (P = 0.004), at both Post-1Y and Post-2Y. Subgroup analysis stratified by type of breast reconstruction revealed significantly higher AS values in patients who underwent prosthetic implant insertion compared to those with autologous reconstruction at all follow-up time points (1st follow-up, P = 0.001; Post-1Y, P < 0.001; and Post-2Y, P < 0.001). A threshold AS of ≥ 1.9 was associated with a 10% predicted risk of developing major complications. The feasibility of an AS generated by a GAN model for predicting both cosmetic outcomes and the likelihood of complications following RT has been successfully validated. Further investigation involving a larger patient cohort is warranted.

The impact of the new ESTRO-ACROP target volume delineation guidelines for postmastectomy radiotherapy after implant-based breast reconstruction on breast complications.

The European Society for Radiotherapy and Oncology-Advisory Committee in Radiation Oncology Practice (ESTRO-ACROP) updated a new target volume delineation guideline for postmastectomy radiotherapy (PMRT) after implant-based reconstruction. This study aimed to evaluate the impact on breast complications with the new guideline compared to the conventional guidelines. In total, 308 patients who underwent PMRT after tissue expander or permanent implant insertion from 2016 to 2021 were included; 184 received PMRT by the new ESTRO-ACROP target delineation (ESTRO-T), and 124 by conventional target delineation (CONV-T). The endpoints were major breast complications (infection, necrosis, dehiscence, capsular contracture, animation deformity, and rupture) requiring re-operation or re-hospitalization and any grade ≥2 breast complications. With a median follow-up of 36.4 months, the cumulative incidence rates of major breast complications at 1, 2, and 3 years were 6.6%, 10.3%, and 12.6% in the ESTRO-T group, and 9.7%, 15.4%, and 16.3% in the CONV-T group; it did not show a significant difference between the groups (p = 0.56). In multivariable analyses, target delineation is not associated with the major complications (sHR = 0.87; p = 0.77). There was no significant difference in any breast complications (3-year incidence, 18.9% vs. 23.3%, respectively; p = 0.56). Symptomatic RT-induced pneumonitis was developed in six (3.2%) and three (2.4%) patients, respectively. One local recurrence occurred in the ESTRO-T group, which was within the ESTRO-target volume. The new ESTRO-ACROP target volume guideline did not demonstrate significant differences in major or any breast complications, although it showed a tendency of reduced complication risks. As the dosimetric benefits of normal organs and comparable oncologic outcomes have been reported, further analyses with long-term follow-up are necessary to evaluate whether it could be connected to better clinical outcomes.

A nationwide study of breast reconstruction after mastectomy in patients with breast cancer receiving postmastectomy radiotherapy: comparison of complications according to radiotherapy fractionation and reconstruction procedures.

BACKGROUND: We examined the patterns of breast reconstruction postmastectomy in breast cancer patients undergoing postmastectomy radiotherapy (PMRT) and compared complications based on radiotherapy fractionation and reconstruction procedures. METHODS: Using National Health Insurance Service (NHIS) data (2015-2020), we analysed 4669 breast cancer patients with PMRT and reconstruction. Using propensity matching, cohorts for hypofractionated fractionation (HF) and conventional fractionation (CF) were created, adjusting for relevant factors and identifying grade ≥3 complications. RESULT: Of 4,669 patients, 30.6% underwent HF and 69.4% CF. The use of HF has increased from 19.4% in 2015 to 41.0% in 2020. Immediate autologous (32.9%) and delayed two-stage implant reconstruction (33.9%) were common. Complication rates for immediate (N = 1286) and delayed two-stage (N = 784) reconstruction were similar between HF and CF groups (5.1% vs. 5.4%, P = 0.803, and 10.5% vs. 10.7%, P = 0.856, respectively) with median follow-ups of 2.5 and 2.6 years. HF showed no increased risk of complications across reconstruction methods. CONCLUSION: A nationwide cohort study revealed no significant difference in complication rates between the HF and CF groups, indicating HF for reconstructed breasts is comparable to CF. However, consultation regarding the fractionation for reconstructed breast cancer patients may still be necessary.

Prediction of Overall Disease Burden in (y)pN1 Breast Cancer Using Knowledge-Based Machine Learning Model.

BACKGROUND: We aimed to construct an expert knowledge-based Bayesian network (BN) model for assessing the overall disease burden (ODB) in (y)pN1 breast cancer patients and compare ODB across arms of ongoing trials. METHODS: Utilizing institutional data and expert surveys, we developed a BN model for (y)pN1 breast cancer. Expert-derived probabilities and disability weights for radiotherapy-related benefit (e.g., 7-year disease-free survival [DFS]) and toxicities were integrated into the model. ODB was defined as the sum of disability weights multiplied by probabilities. In silico predictions were conducted for Alliance A011202, PORT-N1, RAPCHEM, and RT-CHARM trials, comparing ODB, 7-year DFS, and side effects. RESULTS: In the Alliance A011202 trial, 7-year DFS was 80.1% in both arms. Axillary lymph node dissection led to higher clinical lymphedema and ODB compared to sentinel lymph node biopsy with full regional nodal irradiation (RNI). In the PORT-N1 trial, the control arm (whole-breast irradiation [WBI] with RNI or post-mastectomy radiotherapy [PMRT]) had an ODB of 0.254, while the experimental arm (WBI alone or no PMRT) had an ODB of 0.255. In the RAPCHEM trial, the radiotherapy field did not impact the 7-year DFS in ypN1 patients. However, there was a mild ODB increase with a larger irradiation field. In the RT-CHARM trial, we identified factors associated with the major complication rate, which ranged from 18.3% to 22.1%. CONCLUSIONS: The expert knowledge-based BN model predicted ongoing trial outcomes, validating reported results and assumptions. In addition, the model demonstrated the ODB in different arms, with an emphasis on quality of life.

B-Cell-Mediated Immunity Predicts Survival of Patients With Estrogen Receptor-Positive Breast Cancer.

PURPOSE: The estrogen receptor-positive (ER+) breast cancer (BC), which constitutes the majority of BC cases, exhibits highly heterogeneous clinical behavior. To aid precision treatments, we aimed to find molecular subtypes of ER+ BC representing the tumor microenvironment and prognosis. METHODS: We analyzed RNA-seq data of 113 patients with BC and classified them according to the PAM50 intrinsic subtypes using gene expression profiles. Among them, we further focused on 44 patients with luminal-type (ER+) BC for subclassification. The Cancer Genome Atlas (TCGA) data of patients with BC were used as a validation data set to verify the new classification. We estimated the immune cell composition using CIBERSORT and further analyzed its association with clinical or molecular parameters. RESULTS: Principal component analysis clearly divided the patients into two subgroups separately from the luminal A and B classification. The top differentially expressed genes between the subgroups were distinctly characterized by immunoglobulin and B-cell-related genes. We could also cluster a separate cohort of patients with luminal-type BC from TCGA into two subgroups on the basis of the expression of a B-cell-specific gene set, and patients who were predicted to have high B-cell immune activity had better prognoses than other patients. CONCLUSION: Our transcriptomic approach emphasize a molecular phenotype of B-cell immunity in ER+ BC that may help to predict disease prognosis. Although further researches are required, B-cell immunity for patients with ER+ BC may be helpful for identifying patients who are good responders to chemotherapy or immunotherapy.

Assessment of lymph node area coverage with total marrow irradiation and implementation of total marrow and lymphoid irradiation using automated deep learning-based segmentation.

BACKGROUND: Total marrow irradiation (TMI) and total marrow and lymphoid irradiation (TMLI) have the advantages. However, delineating target lesions according to TMI and TMLI plans is labor-intensive and time-consuming. In addition, although the delineation of target lesions between TMI and TMLI differs, the clinical distinction is not clear, and the lymph node (LN) area coverage during TMI remains uncertain. Accordingly, this study calculates the LN area coverage according to the TMI plan. Further, a deep learning-based model for delineating LN areas is trained and evaluated. METHODS: Whole-body regional LN areas were manually contoured in patients treated according to a TMI plan. The dose coverage of the delineated LN areas in the TMI plan was estimated. To train the deep learning model for automatic segmentation, additional whole-body computed tomography data were obtained from other patients. The patients and data were divided into training/validation and test groups and models were developed using the "nnU-NET" framework. The trained models were evaluated using Dice similarity coefficient (DSC), precision, recall, and Hausdorff distance 95 (HD95). The time required to contour and trim predicted results manually using the deep learning model was measured and compared. RESULTS: The dose coverage for LN areas by TMI plan had V100% (the percentage of volume receiving 100% of the prescribed dose), V95%, and V90% median values of 46.0%, 62.1%, and 73.5%, respectively. The lowest V100% values were identified in the inguinal (14.7%), external iliac (21.8%), and para-aortic (42.8%) LNs. The median values of DSC, precision, recall, and HD95 of the trained model were 0.79, 0.83, 0.76, and 2.63, respectively. The time for manual contouring and simply modified predicted contouring were statistically significantly different. CONCLUSIONS: The dose coverage in the inguinal, external iliac, and para-aortic LN areas was suboptimal when treatment is administered according to the TMI plan. This research demonstrates that the automatic delineation of LN areas using deep learning can facilitate the implementation of TMLI.

The Impact of PM2.5 on Radiation-induced Pneumonitis in Patients With Breast Cancer.

BACKGROUND/AIM: Exposure to particulate matter (PM) air pollution is known to adversely affect respiratory disease, but no study has examined its effect on radiation-induced pneumonitis (RIP) in patients with breast cancer. PATIENTS AND METHODS: We conducted a retrospective review of 2,736 patients with breast cancer who received postoperative radiation therapy (RT) between 2017 and 2020 in a single institution. The distance between the PM measurement station and our institution was only 3.43 km. PM data, including PM2.5 and PM10, were retrieved from the open dataset in the official government database. RESULTS: Overall incidence rate of RIP was 1.74%. After adjusting for age, RT technique, regional irradiation, fractionation and boost, the average value of PM2.5 was significantly associated with a higher risk of RIP (p=0.047) when patients received ≥20 fractions of RT. Specifically, PM2.5 ≥35 (µg/m3) showed a significantly higher risk of RIP (p=0.019) in patients with ≥20 fractions of RT. CONCLUSION: This is the first study to reveal the association between PM2.5 and RIP in patients with breast cancer who received 20 fractions or more of postoperative RT. We demonstrated that high PM2.5 levels around the RT institution were associated with RIP, suggesting that reducing PM air pollution may be a modifiable risk factor.

Estimating the risk and benefit of radiation therapy in (y)pN1 stage breast cancer patients: A Bayesian network model incorporating expert knowledge (KROG 22-13).

BACKGROUND: We aimed to evaluate the risk and benefit of (y)pN1 breast cancer patients in a Bayesian network model. METHOD: We developed a Bayesian network (BN) model comprising three parts: pretreatment, intervention, and risk/benefit. The pretreatment part consisted of clinical information from a tertiary medical center. The intervention part regarded the field of radiotherapy. The risk/benefit component encompasses radiotherapy (RT)-related side effects and effectiveness, including factors such as recurrence, cardiac toxicity, lymphedema, and radiation pneumonitis. These factors were evaluated in terms of disability weights and probabilities from a nationwide expert survey. The overall disease burden (ODB) was calculated as the sum of the probability multiplied by the disability weight. A higher value of ODB indicates a greater disease burden for the patient. RESULTS: Among the 58 participants, a BN model utilizing discretization and clustering techniques revealed five distinct clusters. Overall, factors associated with breast reconstruction and RT exhibited high discrepancies (24-34 %), while RT-related side effects demonstrated low discrepancies (3-11 %) among the experts. When incorporating recurrence and RT-related side effects, the mean ODB of (y)pN1 patients was 0.258 (range, 0.244-0.337), with a higher tendency observed in triple-negative breast cancer (TNBC) or mastectomy cases. The ODB for TNBC patients undergoing mastectomy without postmastectomy radiotherapy was 0.327, whereas for non-TNBC patients undergoing breast conserving surgery with RT, the disease burden was 0.251. There was an increasing trend in ODB as the field of RT increased. CONCLUSION: We developed a Bayesian network model based on an expert survey, which helps to understand treatment patterns and enables precise estimations of RT-related risk and benefit in (y)pN1 patients.