Development of deep learning chest X-ray model for cardiac dose prediction in left-sided breast cancer radiotherapy

Abstract

Deep inspiration breath-hold (DIBH) is widely used to reduce the cardiac dose in left-sided breast cancer radiotherapy. This study aimed to develop a deep learning chest X-ray model for cardiac dose prediction to select patients with a potentially high risk of cardiac irradiation and need for DIBH radiotherapy. We used 103 pairs of anteroposterior and lateral chest X-ray data of left-sided breast cancer patients (training cohort: n = 59, validation cohort: n = 19, test cohort: n = 25). All patients underwent breast-conserving surgery followed by DIBH radiotherapy: the treatment plan consisted of three-dimensional, two opposing tangential radiation fields. The prescription dose of the planning target volume was 42.56 Gy in 16 fractions. A convolutional neural network-based regression model was developed to predict the mean heart dose (∆MHD) reduction between free-breathing (MHDFB) and DIBH. The model performance is evaluated as a binary classifier by setting the cutoff value of ∆MHD > 1 Gy. The patient characteristics were as follows: the median (IQR) age was 52 (47–61) years, MHDFB was 1.75 (1.14–2.47) Gy, and ∆MHD was 1.00 (0.52–1.64) Gy. The classification performance of the developed model showed a sensitivity of 85.7%, specificity of 90.9%, a positive predictive value of 92.3%, a negative predictive value of 83.3%, and a diagnostic accuracy of 88.0%. The AUC value of the ROC curve was 0.864. The proposed model could predict ∆MHD in breast radiotherapy, suggesting the potential of a classifier in which patients are more desirable for DIBH.

Mean Heart Dose Prediction Using Parameters of Single-Slice Computed Tomography and Body Mass Index: Machine Learning Approach for Radiotherapy of Left-Sided Breast Cancer of Asian Patients

“…However, the use of this technology by itself, especially in clinical settings, is still limited [ 14 ]. Prior studies have predicted MHD using AI and ML approaches with the aim of selecting patients with a potential risk for cardiac toxicity [ 21 , 22 , 23 , 24 ] and reducing it by performing the DIBH technique [ 21 , 24 ]. In most of these studies, MHD prediction was dependent on CT parameters such as maximum heart distance or cardiac contact distance [ 24 ].…”

“…Prior studies have predicted MHD using AI and ML approaches with the aim of selecting patients with a potential risk for cardiac toxicity [ 21 , 22 , 23 , 24 ] and reducing it by performing the DIBH technique [ 21 , 24 ]. In most of these studies, MHD prediction was dependent on CT parameters such as maximum heart distance or cardiac contact distance [ 24 ]. The prediction of MHD in these studies requires significant time and effort, as it is typically performed after full CT simulation.…”

“…However, the prediction is a time-consuming process, as planning a CT session for DIBH requires an additional 30 min for each patient than undergoing CT with the FB technique [ 21 ]. In contrast, Koide et al also reported that non-CT parameters are promising as predictors of MHD in FB technique, as demonstrated by their use of a CNN based on preoperative chest X-rays of 103 patients with left-sided breast cancer (AUC = 86.4%) [ 24 ]. In both studies, Koide et al attempted to predict ΔMHD, rather than MHD, on FB using the CNN; meanwhile, the performance of the models was evaluated by setting the cut-off value of ∆MHD >100 cGy.…”

“…As for the predictive power of algorithms, DNNs and CNNs, a special type of DNN, are best known for image-related prediction [ 21 , 22 , 24 ]. However, some authors have reported that DNN and CNN were not the best when comparing the predictive power of many algorithms [ 29 , 30 ].…”

A quantitative evaluation method utilizing the homology concept to assess the state of chromatin within the nucleus of lung cancer

“…Once a large imaging dataset and corresponding clinical information are supplied to the machine learning algorithms, common patterns can be identi ed and linked to clinical information. This approach has been used previously for x-ray images because of the normal uniformity of an x-ray image [12][13][14] . However, the images obtained from respiratory cytology present a challenge to the machine learning algorithms, because the patterns of nuclear shape and staining (referred to as the chromatin pattern) are complicated and have a high degree of variability.…”