In this research, the cervical cancer data has been used to build up an ANN model, demonstrating a neural network that has an input layer, hidden layer, and an output layer stated the treatment outcomes for single-channel and tri-channel applicators, activation function identity, and provided the sensitivity and specificity with superior accuracy. The model also described the pseudo-probability, gain, and lift curve with area under the curve (AUC) for single-channel and tri-channel applicator of cervical cancer. The treatment outcomes measured based on survival analysis stated the tri-channel applicator has a higher potential than the single-channel applicator.
Many researchers have used the ANN model to predict the model's based clinical data. Wang et al.[20] proposed an ANN model for traumatic brain injury, demonstrating the prediction of hematoma based on age, bone flop size, glucose level, pupillary response, and the overall accuracy was 73.0%. Azimi et al.[21] reported that the ANN model was established with an accuracy of 96.9% and a better ROC value of 80% for lumbar spinal canal stenosis. Tang et al.[25] used the back prorogation algorithm by artificial neural network for Alzheimer disease screening, resulting in the sensitivity, specificity, and accuracy of 90%, 95%, and 92.50%, respectively. Bottaci et al.[22] suggested that the ANN model for colorectal cancer patients described the sensitivity, specificity, and accuracy of 66%, 88%, and 80%, respectively. Baxt et al.[24] suggested an ANN model for myocardial infection, resulting in the sensitivity and specificity were 97.2% and 96.2%. We found in our study that the model accuracy performance was superior to judge the treatment outcomes used by the applicators in cervical cancer. In the current study, the accuracy was 100%, and 82.4% for the training and testing included AUC=0.961, respectively, in the present study. The sensitivity and specificity were 100% and 100% for training and 87.5%, and 77.8% for testing.
An ANN model has the potential power to predict the risk factor analysis according to the American Society of Anesthesiology (ASA) class > 3 for posterior lumbar spine fusion that has been reported in Kim et al.[23] For cervical cancer, Jaberi et al.[18] proposed image-guided brachytherapy for treatment plan correction of OARs in intra-fraction organ, suggesting the final brachytherapy treatment plan modified based on changed the organ applicators to compensate the target dose controlled at the original level. In chronic lymphocytic leukemia, Aghamaleki et al.[26] proposed an ANN model to detect the molecular biomarker for cancer diagnosis from blood samples. The survival rate studied for gastric cancer patients in Charati et al.[27] The median survival rate was 19±2.04 months at five years, demonstrating an AUC of 94% based on factors such as stage of diseases, metastasis, histology grade, and age of diagnosis. The treatment outcomes based on survival rate were 91.6 % and 89.4 % for Co-60 and Ir-192 at 2-years in the literature in stages Ib2- 111b of cervical cancer.[28] Li et al.[29] have reported that the survival rate for high dose rate brachytherapy for the fletcher group and single-channel group was 80.3 % and 86.3% in cervical cancer at 2-years. In our study, at 2-years, the survival rate was 85% and 95% for the single-channel applicator and tri-channel applicator, respectively. Pang et al.[30] reported an ANN model for the pathological voice by quantities analysis and detection, suggesting the higher accuracy for identification with good clinical information. Li et al.[31] proposed an ANN model to predict the risk factor for heart disease in congenital heart disease, suggesting the sensitivity, specificity was 87% and 90% for the training set. The AUC value of training and test set were 0.87 and 0.97, respectively. Kuang et al.[32] proposed an ANN model for Alzheimer's disease, describing the sensitivity, specificity, and AUC were 82.11±0.42%, 75.26±0.86%, and 92.08±0.12%, respectively, included accuracy 89.52±0.36%. In our study, the AUC value was 0.961.
Rajković et al.[17] suggested an ANN model for the treatment of prostate carcinoma, resulting in the therapy dose (TD) of 47.3 Gy and coverage index (CI100%) of 1.4 for the low-risk group and TD of 50.4 Gy and CI100% 1.6 for the high-risk group. In this research, we treated better therapy doses for cervical cancer patients to build up the ANNs model. The EBRT was 45 (40-50) cGy after chemotherapy. During the period of brachytherapy, the rectum dose and bladder dose were in the following: Rectum Dose 3.71 (2.29-5.14) Gy; Bladder Dose 2 (1.94-5.83) Gy and Prescribe Dose 6.87 (6-7) Gy. We found Tumor Area Dose 6.74 (5.64-7) Gy and Dose Volume 113.6 (58.2-469.25) cm3.
There are some limitations. The ANNs can identify the complex and non-linear relationship between independent and dependent variables and detect all possible interactions for all predictors.[33] The ANNs have some disadvantages. The 'Black Box' cannot have explained the odd ratio that identifies the direction and magnitude of the effect of each variable like LR.[34] The ANNs model is prone to adjust the overfitting data that the model is not perfect for generalization to the external data.[33] The optimization problem of the ANNs model is complex, such as training times, several nodes, regulations, and layers to proceed optimally the outcomes.[35]