Multimodal radiotherapy dose prediction using a multi-task deep learning model.

BACKGROUND: In radiation therapy (RT), accelerated partial breast irradiation (APBI) has emerged as an increasingly preferred treatment modality over conventional whole breast irradiation due to its targeted dose delivery and shorter course of treatment. APBI can be delivered through various modalities including Cobalt-60-based systems and linear accelerators with C-arm, O-ring, or robotic arm design. Each modality possesses distinct features, such as beam energy or the degrees of freedom in treatment planning, which influence their respective dose distributions. These modality-specific considerations emphasize the need for a quantitative approach in determining the optimal dose delivery modality on a patient-specific basis. However, manually generating treatment plans for each modality across every patient is time-consuming and clinically impractical. PURPOSE: We aim to develop an efficient and personalized approach for determining the optimal RT modality for APBI by training predictive models using two different deep learning-based convolutional neural networks. The baseline network performs a single-task (ST), predicting dose for a single modality. Our proposed multi-task (MT) network, which is capable of leveraging shared information among different tasks, can concurrently predict dose distributions for various RT modalities. Utilizing patient-specific input data, such as a patient's computed tomography (CT) scan and treatment protocol dosimetric goals, the MT model predicts patient-specific dose distributions across all trained modalities. These dose distributions provide patients and clinicians quantitative insights, facilitating informed and personalized modality comparison prior to treatment planning. METHODS: The dataset, comprising 28 APBI patients and their 92 treatment plans, was partitioned into training, validation, and test subsets. Eight patients were dedicated to the test subset, leaving 68 treatment plans across 20 patients to divide between the training and validation subsets. ST models were trained for each modality, and one MT model was trained to predict doses for all modalities simultaneously. Model performance was evaluated across the test dataset in terms of Mean Absolute Percent Error (MAPE). We conducted statistical analysis of model performance using the two-tailed Wilcoxon signed-rank test. RESULTS: Training times for five ST models ranged from 255 to 430 min per modality, totaling 1925 min, while the MT model required 2384 min. MT model prediction required an average of 1.82 s per patient, compared to ST model predictions at 0.93 s per modality. The MT model yielded MAPE of 1.1033 ± 0.3627% as opposed to the collective MAPE of 1.2386 ± 0.3872% from ST models, and the differences were statistically significant (p = 0.0003, 95% confidence interval = [-0.0865, -0.0712]). CONCLUSION: Our study highlights the potential benefits of a MT learning framework in predicting RT dose distributions across various modalities without notable compromises. This MT architecture approach offers several advantages, such as flexibility, scalability, and streamlined model management, making it an appealing solution for clinical deployment. With such a MT model, patients can make more informed treatment decisions, physicians gain more quantitative insight for pre-treatment decision-making, and clinics can better optimize resource allocation. With our proposed goal array and MT framework, we aim to expand this work to a site-agnostic dose prediction model, enhancing its generalizability and applicability.

Master Breast Radiation Planning: Simple Guide for Radiation Oncology Residents.

This article focuses on various aspects of breast radiation treatment planning, from simulation to field design. It covers the most common techniques including tangents, mono isocentric, dual isocentric, electron-photon match, and VMAT. This can serve as a guide for radiation oncology residents and medical students to advance their understanding of key aspects of breast radiation treatment and planning processes.

Omission of Radiotherapy After Breast-Conserving Surgery for Women With Breast Cancer With Low Clinical and Genomic Risk: 5-Year Outcomes of IDEA.

PURPOSE: Multiple studies have shown a low risk of ipsilateral breast events (IBEs) or other recurrences for selected patients age 65-70 years or older with stage I breast cancers treated with breast-conserving surgery (BCS) and endocrine therapy (ET) without adjuvant radiotherapy. We sought to evaluate whether younger postmenopausal patients could also be successfully treated without radiation therapy, adding a genomic assay to classic selection factors. METHODS: Postmenopausal patients age 50-69 years with pT1N0 unifocal invasive breast cancer with margins ≥2 mm after BCS whose tumors were estrogen receptor-positive, progesterone receptor-positive, and human epidermal growth factor receptor 2-negative with Oncotype DX 21-gene recurrence score ≤18 were prospectively enrolled in a single-arm trial of radiotherapy omission if they consented to take at least 5 years of ET. The primary end point was the rate of locoregional recurrence 5 years after BCS. RESULTS: Between June 2015 and October 2018, 200 eligible patients were enrolled. Among the 186 patients with clinical follow-up of at least 56 months, overall and breast cancer-specific survival rates at 5 years were both 100%. The 5-year freedom from any recurrence was 99% (95% CI, 96 to 100). Crude rates of IBEs for the entire follow-up period for patients age 50-59 years and age 60-69 years were 3.3% (2/60) and 3.6% (5/140), respectively; crude rates of overall recurrence were 5.0% (3/60) and 3.6% (5/140), respectively. CONCLUSION: This trial achieved a very low risk of recurrence using a genomic assay in combination with classic clinical and biologic features for treatment selection, including postmenopausal patients younger than 60 years. Long-term follow-up of this trial and others will help determine whether the option of avoiding initial radiotherapy can be offered to a broader group of women than current guidelines recommend.

Revisiting Exclusion of Prior Cancer in Clinical Trials of Male Breast Cancer.

Background: Eligibility criteria for cancer clinical trials present challenges to enrollment. Many trials exclude patients with a prior cancer. This common practice may be especially detrimental to trials of rare cancers, such as male breast cancer, that struggle to accrue adequate numbers of participants. Objectives: To estimate prevalence of prior cancer among men newly diagnosed with breast cancer and describe characteristics of men with prior cancer compared to those without. Methods: We identified men diagnosed with breast cancer between 2011-2015 using population-based data from National Cancer Institute's Surveillance, Epidemiology, and End Results program of cancer registries. We used sequence number and diagnosis year to identify cancers diagnosed prior to breast cancer (inclusive of prior breast, different, and unknown types of cancer). We compared sociodemographic, tumor, and treatment characteristics of men with and without prior cancer using chi-square tests. Results: Among 2317 men, nearly one quarter (24.3%) had any prior cancer, and the majority (58.7%) of these were of a different cancer type. A higher proportion of men with a prior cancer of a different type were older, had smaller (≤ 2 cm) breast tumors, were diagnosed with stage 0-1 breast cancer, and did not receive surgery compared to men without any prior cancer; there were no statistically significant differences by race and ethnicity, county median income, hormone receptor status, or surgery type. Conclusion: Given prevalence of prior cancer in this rare and understudied population of men diagnosed with breast cancer, including men with prior cancer in clinical trials may improve accrual.

Real World and Public Health Perspectives of Intraoperative Radiotherapy in Early-Stage Breast Cancer: A Multidisciplinary Analysis Beyond the Statistical Facts.

Breast conservation therapy (BCT) (usually a lumpectomy plus radiotherapy (RT)) has become a standard alternative to radical mastectomy in early-stage breast cancers with equal, if not higher, survival rates. The established standard of the RT component of the BCT had been about six weeks of Monday through Friday external beam RT to the whole breast (WBRT). Recent clinical trials have shown that partial breast radiation therapy (PBRT) to the region surrounding the lumpectomy cavity with shorter courses can result in equal local control, survival, and slightly improved cosmetic outcomes. Intraoperative RT (IORT) wherein RT is administered at the time of operation for BCT to the lumpectomy cavity as a single-fraction RT is also considered PBRT. The advantage of IORT is that weeks of RT are avoided. However, the role of IORT as part of BCT has been controversial. The extreme views go from "I will not recommend to anyone" to "I can recommend to all early-stage favorable patients." These divergent views are due to difficulty in interpreting the clinical trial results. There are two modalities of delivering IORT, namely, the use of low-energy 50 kV beams or electron beams. There are several retrospective, prospective, and two randomized clinical trials comparing IORT versus WBRT. Yet, the opinions are divided. In this paper, we try to bring clarity and consensus from a highly broad-based multidisciplinary team approach. The multidisciplinary team included breast surgeons, radiation oncologists, medical physicists, biostatisticians, public health experts, nurse practitioners, and medical oncologists. We show that there is a need to more carefully interpret and differentiate the data based on electron versus low-dose X-ray modalities; the randomized study results have to be extremely carefully dissected from biostatistical points of view; the importance of the involvement of patients and families in the decision making in a very transparent and informed manner needs to be emphasized; and the compromise some women may be willing to accept between 2-4% potential increase in local recurrence (as interpreted by some of the investigators in IORT randomized studies) versus mastectomy. We conclude that, ultimately, the choice should be that of women with detailed facts of the pros and cons of all options being presented to them from the angle of patient/family-focused care. Although the guidelines of various professional societies can be helpful, they are only guidelines. The participation of women in IORT clinical trials is still needed, and as genome-based and omics-based fine-tuning of prognostic fingerprints evolve, the current guidelines need to be revisited. Finally, the use of IORT can help rural, socioeconomically, and infrastructure-deprived populations and geographic regions as the convenience of single-fraction RT and the possibility of breast preservation are likely to encourage more women to choose BCT than mastectomy. This option can also likely lead to more women choosing to get screened for breast cancer, thus enabling the diagnosis of breast cancer at an earlier stage and improving the survival outcomes.

A deep learning approach for automatic delineation of clinical target volume in stereotactic partial breast irradiation (S-PBI).

Accurate and efficient delineation of the clinical target volume (CTV) is of utmost significance in post-operative breast cancer radiotherapy. However, CTV delineation is challenging as the exact extent of microscopic disease encompassed by CTV is not visualizable in radiological images and remains uncertain. We proposed to mimic physicians' contouring practice for CTV segmentation in stereotactic partial breast irradiation (S-PBI) where CTV is derived from tumor bed volume (TBV) via a margin expansion followed by correcting the extensions for anatomical barriers of tumor invasion (e.g. skin, chest wall). We proposed a deep-learning model, where CT images and the corresponding TBV masks formed a multi-channel input for a 3D U-Net based architecture. The design guided the model to encode the location-related image features and directed the network to focus on TBV to initiate CTV segmentation. Gradient weighted class activation map (Grad-CAM) visualizations of the model predictions revealed that the extension rules and geometric/anatomical boundaries were learnt during model training to assist the network to limit the expansion to a certain distance from the chest wall and the skin. We retrospectively collected 175 prone CT images from 35 post-operative breast cancer patients who received 5-fraction partial breast irradiation regimen on GammaPod. The 35 patients were randomly split into training (25), validation (5) and test (5) sets. Our model achieved mean (standard deviation) of 0.94 (±0.02), 2.46 (±0.5) mm, and 0.53 (±0.14) mm for Dice similarity coefficient, 95th percentile Hausdorff distance, and average symmetric surface distance respectively on the test set. The results are promising for improving the efficiency and accuracy of CTV delineation during on-line treatment planning procedure.

Aspirin use is associated with improvement in distant metastases outcome in patients with residual disease after neoadjuvant chemotherapy.

PURPOSE: Aspirin (ASA) use has been correlated with improved outcomes in high-risk patients at risk for distant metastases. Breast cancer (BC) patients with residual disease, particularly nodal disease (ypN +) after neoadjuvant chemotherapy (NAC), are high-risk patients portending worse outcomes. We hypothesized that ASA use can reduce distant metastases and improve outcomes in these patients. METHODS: Patients at our institutions from 2005 to 2018, with BC who did not achieve complete response (pCR) after NAC were reviewed (IRB protocol STU- 052012-019). Data, including evidence of ASA use, and clinico-pathologic parameters were analyzed. Survival outcomes were obtained (Kaplan Meier analysis) and univariate (UVA) and multivariable (MVA) Cox proportional hazards regression analyses were performed. RESULTS: 637 did not achieve pCR (ypN+ = 422). 138 were ASA users. Median follow-up for the control and ASA group were 3.8 (IQR 2.2-6.3) and 3.8 (IQR 2.5-6.4) years, respectively. Majority were stage II/III. 387 were hormone receptor positive, 191 HER2 +, and 157 triple negative. On UVA, ASA use, PR status, pathologic and clinical stage showed significance for DMFS, and disease-free survival (DFS). On MVA, ASA use associated with improved 5-year DFS (p = .01, 87.0% vs 79.6%, adjusted HR = 0.48) and improved 5-year DMFS (p = .04, 92.8% vs 89.2%, adjusted HR = 0.57). In the ypN + patients, ASA use associated with improved 5-year DMFS (p = .008, 85.7% vs 70.7%, adjusted HR = 0.43) and DFS (p = .02, 86.8% vs 74.3%, adjusted HR = 0.48). CONCLUSION: For non-responders, particularly ypN + patients, ASA use associated with improved outcome. These hypotheses-generating results suggest for development of prospective clinical trials of augmented ASA use in selected very high-risk BC patients.

Aspirin Use Is Associated With Improved Outcomes in Inflammatory Breast Cancer Patients.

PURPOSE: Inflammatory breast cancer (IBC) is the most aggressive form of breast cancer and has a high propensity for distant metastases. Our previous data suggested that aspirin (acetylsalicylic acid, ASA) use may be associated with reduced risk of distant metastases in aggressive breast cancer; however, there are no reported studies on the potential benefit of ASA use in patients with IBC. METHODS: Data from patients with non-metastatic IBC treated between 2000-2017 at two institutions, were reviewed. Overall survival (OS), disease-free survival (DFS), and distant metastasis-free survival (DMFS) were performed using Kaplan-Meier analysis. Univariate and multivariable logistic regression models were used to identify significant associated factors. RESULTS: Of 59 patients meeting the criteria for analysis and available for review, 14 ASA users were identified. ASA users demonstrated increased OS (p = 0.03) and DMFS (p = 0.02), with 5-year OS and DMFS of 92% (p = 0.01) and 85% (p = 0.01) compared to 51% and 43%, respectively, for non-ASA users. In univariate analysis, pT stage, pN stage, and ASA use were significantly correlated (p < 0.05) with OS and DFS. On multivariable analysis, ASA use (hazard ratio [HR], 0.11; 95% confidence interval [CI], 0.01-0.8) and lymph node stage (HR, 5.9; 95% CI, 1.4-25.9) remained significant for OS and DFS ASA use (HR, 0.13; 95% CI, 0.03-0.56) and lymph node stage (HR, 5.6; 95% CI, 1.9-16.4). CONCLUSION: ASA use during remission was associated with significantly improved OS and DMFS in patients with IBC. These results suggest that ASA may provide survival benefits to patients with IBC. Prospective clinical trials of ASA use in patients with high-risk IBC in remission should be considered.

Stability and reproducibility comparisons between deep inspiration breath-hold techniques for left-sided breast cancer patients: A prospective study.

PURPOSE: Deep inspiration breath-hold (DIBH) is crucial in reducing the lung and cardiac dose for treatment of left-sided breast cancer. We compared the stability and reproducibility of two DIBH techniques: Active Breathing Coordinator (ABC) and VisionRT (VRT). MATERIALS AND METHODS: We examined intra- and inter-fraction positional variation of the left lung. Eight left-sided breast cancer patients were monitored with electronic portal imaging during breath-hold (BH) at every fraction. For each patient, half of the fractions were treated using ABC and the other half with VRT, with an equal amount starting with either ABC or VRT. The lung in each portal image was delineated, and the variation of its area was evaluated. Intrafraction stability was evaluated as the mean coefficient of variation (CV) of the lung area for the supraclavicular (SCV) and left lateral (LLat) field over the course of treatment. Reproducibility was the CV for the first image of each fraction. Daily session time and total imaging monitor units (MU) used in patient positioning were recorded. RESULTS: The mean intrafraction stability across all patients for the LLat field was 1.3 ± 0.7% and 1.5 ± 0.9% for VRT and ABC, respectively. Similarly, this was 1.5 ± 0.7% and 1.6 ± 0.8% for VRT and ABC, respectively, for the SCV field. The mean interfraction reproducibility for the LLat field was 11.0 ± 3.4% and 14.9 ± 6.0% for VRT and ABC, respectively. Similarly, this was 13.0 ± 2.5% and 14.8 ± 9% for VRT and ABC, respectively, for the SCV. No difference was observed in the number of verification images required for either technique. CONCLUSIONS: The stability and reproducibility were found to be comparable between ABC and VRT. ABC can have larger interfractional variation with less feedback to the treating therapist compared to VRT as shown in the increase in geometric misses at the matchline.

Towards precision radiation oncology: endocrine therapy response as a biomarker for personalization of breast radiotherapy.

Targeted therapies, such as endocrine therapies (ET), can exert selective pressure on cancer cells and promote adaptations that confer treatment resistance. In this study, we show that ET resistance in breast cancer drives radiation resistance through reprogramming of DNA repair pathways. We also show that pharmacological bromodomain and extraterminal domain inhibition reverses pathological DNA repair reprogramming in ET-resistant breast tumors and overcomes resistance to radiation therapy.

Feasibility and efficacy of active breathing coordinator assisted deep inspiration breath hold technique for treatment of locally advanced breast cancer.

BACKGROUND: Active breathing coordinator (ABC)-assisted deep inspiration breath hold (DIBH) is an important organ sparing radiation therapy (RT) technique for left-sided breast cancer patients. Patients with advanced breast cancer undergoing chest wall and regional nodal irradiation often require a field matching technique. While field matching has been demonstrated to be safe and effective in free breathing patients, its safety and accuracy in DIBH/ABC use has not been previously reported. PURPOSE: To report the accuracy, feasibility, and safety of field matching with ABC/DIBH for patients receiving breast/chest wall irradiation with nodal irradiation using a three-field technique. METHODS: From December 2012 to May 2018, breast cancer patients undergoing ABC/DIBH-based RT at a single institution were reviewed. For each fraction, the amount of overlap/gap between the supraclavicular and the tangential field were measured and recorded. Patient characteristics, including acute and delayed skin toxicities, were analyzed. RESULTS: A total of 202 patients utilized ABC/DIBH and 4973 fractions had gap/overlap measurements available for analysis. The average gap/overlap measured at junction was 0.28 mm ± 0.99 mm. A total of 72% of fractions had no measurable gap/overlap (0 mm), while 5.6% had an overlap and 22.7% a gap. There was no significant trend for worsening or improvement of gap/overlap measurements with increasing fraction number per patient. OSLD measurements were compared to the planned dose. The median dose 1 cm above the junction was 106% ± 7% of planned dose (range 94%-116%). One centimeter below the junction, the median dose was 114% ± 11% of planned dose (range 95%-131%). At the junction, the median dose was 106% ± 16.3% of planned dose (range 86%-131%). Acute skin toxicity was similar to historically reported values (grade 3, 5.4%, grade 4, 0%). CONCLUSION: ABC-assisted DIBH is a safe and technically feasible method of delivering RT in the setting of complex matching field technique for breast and regional nodal treatments.

On the feasibility of improved target coverage without compromising organs at risk using online adaptive stereotactic partial breast irradiation (A-SPBI).

PURPOSE: Describe an early-adopting institution's experience with online adaptive radiation for stereotactic partial breast irradiation. METHODS AND MATERIALS: Retrospective review of 22 women treated between May 2021 and March 2022 with adaptive stereotactic partial breast irradiation. A total of 106 of 110 fractions were evaluated for dosimetric changes in target coverage and organ-at-risk (OAR) dose. Patient set up with stereotactic wooden frame and adapted per fraction. Treatment and planning times were collected prospectively by radiation therapists. RESULTS: Scheduled PTV30 Gy was <95% in 72.1% and <90% in 38.5% of fractions, and both PTV and CTV coverage were improved significantly after adaption, and 83.7% of fractions were delivered as adapted per physician choice. There was no difference in OAR coverage. Average adaptive treatment planning took 15 min and average time-on-couch was 34.4 min. CONCLUSIONS: Adaptive stereotactic breast irradiation resulted in improved target coverage with equivalent dosing to OARs in an efficient and tolerated treatment time. Improved target coverage allowed for decreased PTV margins compared to prior trial protocols that may improve acute and late toxicities.

Stereotactic Partial Breast Irradiation: What Does the Future Hold?

Breast irradiation has evolved significantly over the last several decades. Accelerated partial breast and stereotactic breast irradiation have evolved as strategies to reduce irradiated volumes, preserve appropriate oncologic control, and improve cosmetic outcome. The sequencing and/or combination of stereotactic partial breast irradiation with novel systemic agents is of great interest to the oncologic community. Here we explore the landscape of modern trials and opine on the future of partial breast irradiation.

Potential Impact of Revised NCI Eligibility Criteria Guidance: Prior Malignancy Exclusion in Breast Cancer Clinical Trials.

BACKGROUND: Many individuals with cancer have survived a prior cancer and for this reason may have been excluded from clinical trials. Recent NCI guidance recommends including these individuals, especially when the risk of the prior malignancy interfering with either safety or efficacy endpoints is very low. Using breast cancer as an example, we determined the potential effect this policy change may have on clinical trial accrual. PATIENTS AND METHODS: We reviewed protocols of NCI-sponsored breast cancer clinical trials activated in 1991 through 2016. We quantified prevalence of prior cancer-related exclusion criteria and assessed the association with trial characteristics using Fisher's exact tests. Using SEER data, we estimated the prevalence and timing of prior primary (nonbreast) cancer diagnoses among patients with breast cancer. RESULTS: Among 87 clinical trials (total target enrollment, 137,253 patients), 77% excluded individuals with prior cancer, most commonly (79%) within the preceding 5 years. Among trials with radiographic response or toxicity endpoints, 69% excluded prior cancer. In SEER data, the prevalence of a prior (nonbreast) cancer diagnosis ranged from 5.7% to 7.7%, depending on breast cancer stage, of which 39% occurred within 5 years of the incident breast cancer. For trials excluding prior cancer, the estimated proportion of patients excluded for this reason ranged from 1.3% to 5.8%, with the estimated number of excluded patients ranging from 1 to 288. CONCLUSIONS: More than three-fourths of NCI-sponsored breast cancer clinical trials exclude patients with prior cancer, including almost 70% of trials with response or toxicity endpoints. Given that >5% of patients with breast cancer have a history of prior cancer, in large phase III trials this practice may exclude hundreds of patients. Following recent NCI eligibility guidance, the inclusion of patients with prior cancer on breast cancer trials may have a meaningful impact on accrual.

A How-To Compendium for GammaPod Treatments, Clinical Workflow, and Clinical Program at an Early Adopting Institution.

In 2019, our institution became the second in the world to go live with GammaPod (Xcision Medical Systems, LLC, Columbia, MD), a device dedicated for stereotactic radiation therapy of breast cancer, with breast immobilization, real-time imaging, and highly-conformal dosimetry. At our institution, GammaPod is used for 5-fraction adjuvant partial breast irradiation, single-fraction tumor cavity boost before whole-breast irradiation, single-fraction preoperative radiation, and (in poor surgical candidates), single-fraction definitive radiation. Here, we describe our workflow, observed procedure step times, and homegrown techniques for improved efficiency in our institutional experience of 93 patients treated between 2019 and 2021.

Preliminary Results of Multi-Institutional Phase 1 Dose Escalation Trial Using Single-Fraction Stereotactic Partial Breast Irradiation for Early Stage Breast Cancer.

PURPOSE: We report on our early experience of our prospective multicenter phase 1 dose- escalation study of single-fraction stereotactic partial breast irradiation (S-PBI) for early stage breast cancer after partial mastectomy using a robotic stereotactic radiation system. METHODS AND MATERIALS: Thirty women with in situ or invasive breast cancer stage 0, I, or II with tumor size <3 cm treated with lumpectomy were enrolled in this phase 1 single-fraction S-PBI dose-escalation trial. Women received either 22.5, 26.5, or 30 Gy in a single fraction using a robotic stereotactic radiation system. The primary outcome was to reach tumoricidal dose of 30 Gy in a single fraction to the lumpectomy cavity without exceeding the maximum tolerated dose. Secondary outcomes were to determine dose-limiting toxicity and cosmesis. Tertiary goals were ipsilateral breast recurrence rate, distant disease-free interval, recurrence-free survival, and overall survival. RESULTS: From June 2016 to January 2021, 11, 8, and 10 patients were treated to doses of 22.5, 26.5, or 30 Gy in a single fraction, respectively, with median follow-up being 47.9, 25.1, and 16.2 months. No patients experienced acute (<90 days) grade 3 or higher treatment-related toxicity, and maximum tolerated dose was not reached. There were 2 delayed grade 3 toxicities. Four patients (13.8%) developed fat necrosis across all 3 cohorts, which compares favorably with results from other PBI trials. No dose cohort had a statistically significant cosmetic detriment from baseline to 12 months or 24 months follow-up by patient- or physician-reported global cosmetic scores. There were no reports of disease recurrence. CONCLUSIONS: This phase 1 trial demonstrates that S-PBI can be used to safely escalate dose to 30 Gy in a single fraction with low toxicity and without detriment in cosmesis relative to baseline.

Saliency-guided deep learning network for automatic tumor bed volume delineation in post-operative breast irradiation.

Efficient, reliable and reproducible target volume delineation is a key step in the effective planning of breast radiotherapy. However, post-operative breast target delineation is challenging as the contrast between the tumor bed volume (TBV) and normal breast tissue is relatively low in CT images. In this study, we propose to mimic the marker-guidance procedure in manual target delineation. We developed a saliency-based deep learning segmentation (SDL-Seg) algorithm for accurate TBV segmentation in post-operative breast irradiation. The SDL-Seg algorithm incorporates saliency information in the form of markers' location cues into a U-Net model. The design forces the model to encode the location-related features, which underscores regions with high saliency levels and suppresses low saliency regions. The saliency maps were generated by identifying markers on CT images. Markers' location were then converted to probability maps using a distance transformation coupled with a Gaussian filter. Subsequently, the CT images and the corresponding saliency maps formed a multi-channel input for the SDL-Seg network. Our in-house dataset was comprised of 145 prone CT images from 29 post-operative breast cancer patients, who received 5-fraction partial breast irradiation (PBI) regimen on GammaPod. The 29 patients were randomly split into training (19), validation (5) and test (5) sets. The performance of the proposed method was compared against basic U-Net. Our model achieved mean (standard deviation) of 76.4(±2.7) %, 6.76(±1.83) mm, and 1.9(±0.66) mm for Dice similarity coefficient, 95 percentile Hausdorff distance, and average symmetric surface distance respectively on the test set with computation time of below 11 seconds per one CT volume. SDL-Seg showed superior performance relative to basic U-Net for all the evaluation metrics while preserving low computation cost. The findings demonstrate that SDL-Seg is a promising approach for improving the efficiency and accuracy of the on-line treatment planning procedure of PBI, such as GammaPod based PBI.