RESUMEN

BACKGROUND: The reliable and efficient estimation of uncertainty in artificial intelligence (AI) models poses an ongoing challenge in many fields such as radiation therapy. AI models are intended to automate manual steps involved in the treatment planning workflow. We focus in this study on dose prediction models that predict an optimal dose trade-off for each new patient for a specific treatment modality. They can guide physicians in the optimization, be part of automatic treatment plan generation or support decision in treatment indication. Most common uncertainty estimation methods are based on Bayesian approximations, like Monte Carlo dropout (MCDO) or Deep ensembling (DE). These two techniques, however, have a high inference time (i.e., require multiple inference passes) and might not work for detecting out-of-distribution (OOD) data (i.e., overlapping uncertainty estimate for in-distribution (ID) and OOD). PURPOSE: In this study, we present a direct uncertainty estimation method and apply it for a dose prediction U-Net architecture. It can be used to flag OOD data and give information on the quality of the dose prediction. METHODS: Our method consists in the addition of a branch decoding from the bottleneck which reconstructs the CT scan given as input. The input reconstruction error can be used as a surrogate of the model uncertainty. For the proof-of-concept, our method is applied to proton therapy dose prediction in head and neck cancer patients. A dataset of 60 oropharyngeal patients was used to train the network using a nested cross-validation approach with 11 folds (training: 50 patients, validation: 5 patients, test: 5 patients). For the OOD experiment, we used 10 extra patients with a different head and neck sub-location. Accuracy, time-gain, and OOD detection are analyzed for our method in this particular application and compared with the popular MCDO and DE. RESULTS: The additional branch did not reduce the accuracy of the dose prediction model. The median absolute error is close to zero for the target volumes and less than 1% of the dose prescription for organs at risk. Our input reconstruction method showed a higher Pearson correlation coefficient with the prediction error (0.620) than DE (0.447) and MCDO (between 0.599 and 0.612). Moreover, our method allows an easier identification of OOD (no overlap for ID and OOD data and a Z-score of 34.05). The uncertainty is estimated simultaneously to the regression task, therefore requires less time and computational resources. CONCLUSIONS: This study shows that the error in the CT scan reconstruction can be used as a surrogate of the uncertainty of the model. The Pearson correlation coefficient with the dose prediction error is slightly higher than state-of-the-art techniques. OOD data can be more easily detected and the uncertainty metric is computed simultaneously to the regression task, therefore faster than MCDO or DE. The code and pretrained model are available on the gitlab repository: https://gitlab.com/ai4miro/ct-reconstruction-for-uncertainty-quatification-of-hdunet.

RESUMEN

BACKGROUND AND PURPOSE: Deep inspiration breath-hold (DIBH) protects critical organs-at-risk (OARs) for adjuvant breast radiotherapy. Guidance systems e.g. surface guided radiation therapy (SGRT) improve the positional breast reproducibility and stability during DIBH. In parallel, OARs sparing with DIBH is enhanced through different techniques e.g. prone position, continuous positive airway pressure (CPAP). By inducing repeated DIBH with the same level of positive pressure, mechanically-assisted and non-invasive ventilation (MANIV) could potentially combine these DIBH optimizations. MATERIALS AND METHODS: We conducted a randomized, open-label, multicenter and single-institution non-inferiority trial. Sixty-six patients eligible for adjuvant left whole-breast radiotherapy in supine position were equally assigned between mechanically-induced DIBH (MANIV-DIBH) and voluntary DIBH guided by SGRT (sDIBH). The co-primary endpoints were positional breast stability and reproducibility with a non-inferiority margin of 1 mm. Secondary endpoints were tolerance assessed daily via validated scales, treatment time, dose to OARs and their inter-fraction positional reproducibility. RESULTS: Differences between both arms for positional breast reproducibility and stability occurred at a sub-millimetric level (p < 0.001 for non-inferiority). The left anterior descending artery near-max dose (14,6 ± 12,0 Gy vs. 7,7 ± 7,1 Gy, p = 0,018) and mean dose (5,0 ± 3,5 Gy vs. 3,0 ± 2,0 Gy, p = 0,009) were improved with MANIV-DIBH. The same applied for the V5Gy of the left ventricle (2,4 ± 4,1 % vs. 0,8 ± 1,6 %, p = 0,001) as well as for the left lung V20Gy (11,4 ± 2,8 % vs. 9,7 ± 2,7 %, p = 0,019) and V30Gy (8,0 ± 2,6 % vs. 6,5 ± 2,3 %, p = 0,0018). Better heart's inter-fraction positional reproducibility was observed with MANIV-DIBH. Tolerance and treatment time were similar. CONCLUSION: Mechanical ventilation provides the same target irradiation accuracy as with SGRT while better protecting and repositioning OARs.

Asunto(s)

RESUMEN

PURPOSE: This prospective, nonrandomized, interventional phase 1-2 study investigated the individualization of elective node irradiation in clinically N0 head and neck squamous cell carcinoma by sentinel lymph node (SLN) mapping with single-photon emission computed tomography/computed tomography (SPECT/CT) and its impact on tumor control and radiation-related toxicity. METHODS AND MATERIALS: Forty-four patients with clinically N0 head and neck squamous cell carcinoma treated with definitive (chemo-)radiation therapy were imaged with SPECT/CT after 99mTc nanocolloid injection around the tumor. The neck levels containing up to the 4 hottest SLNs were selected for prophylactic irradiation. A comparative virtual planning was performed with the selection of neck levels based on the current international guidelines. Regional control was monitored as a function of the selected volume. Dosimetric data for the organs at risk were compared between the plans. Normal tissue complication probability (NTCP) rates were derived for xerostomia, dysphagia, and hypothyroidism to predict the clinical benefit and correlated to quality-of-life (QoL) assessments at 6 months. RESULTS: Sixteen percent of patients presented unpredicted lymphatic drainage, and 48% drained unilaterally. The nodal clinical target volume based on lymphoscintigraphy was smaller than the nodal clinical target volume based on international guidelines by a factor of 2 (P < .0001). After a median follow-up of 46 months, only 1 patient experienced a regional relapse in a nonirradiated area. Significant median dose reductions to organs at risk were observed, particularly to contralateral salivary glands in patients with unilateral drainage (14.6-28.1 Gy) and to the thyroid gland in all patients (22.4-48.9 Gy). Median NTCP reductions were observed for xerostomia (0.3% to 13.7%), dysphagia (1.7% to 10.8%), and hypothyroidism (14.0% to 36.1%). QoL at 6 months was improved, particularly in patients irradiated unilaterally. CONCLUSIONS: Neck SLN mapping with SPECT/CT individualizes and reduces the elective nodal target volumes without compromising the regional control. The NTCP rates were reduced and favorable QoL were observed in all patients, particularly in the case of unilateral irradiation.

Asunto(s)

RESUMEN

Despite a better understanding in head and neck tumors pathogenesis as well as improvements in radiotherapy and surgery, locally advanced head and neck squamous cell carcinoma (HNSCC) remains of poor prognosis. One promising target is the epidermal growth factor receptor (EGFR), which is overexpressed in the majority of HNSCC and is associated to tumor progression and resistance to treatment. However, in several clinical trials, the combination of EGFR inhibitors with chemotherapy and/or radiotherapy generates moderate results. In this study, we investigated the anti-tumor activity of afatinib, an irreversible pan-EGFR inhibitor, combined to cisplatin in different schedules of exposure. For that, we used two human EGFR wild-type HNSCC cell lines and we evaluated the cytotoxicity of the two drugs combined in different sequences. The efficiency of each strategy was assessed by evaluating the effects on cell cycle distribution, DNA damage, cell death and downstream pathways of ErbB family receptors. We demonstrated that cisplatin treatment followed by afatinib exposure displayed more cytotoxic effects than the opposite timing or than simultaneous association. This higher anticancer activity is probably due to afatinib-induced cell cycle arrest, which prevents the repair of cisplatin-induced DNA damage and promotes cell death by various mechanisms including apoptosis. These data suggest the importance of an appropriate timing administration between an EGFR inhibitor and a conventional chemotherapy in order to obtain the best clinical benefit for patients with a head and neck cancer.