Enhancing Classification of Primary and Recurrent Epithelial Ovarian Cancer

Epithelial Ovarian Carcinoma (EOC) is one of the fatal cancers, with intricate molecular features that impact its diagnosis and treatment within the female reproductive system worldwide.EOC remains one of the most lethal gynaecological malignancies, primarily due to the molecular heterogeneity in its occurrence, which complicates the classification of Primary Tumours (PT) and recurrent Tumours (RT) using traditional methods.The critical challenge lies in the size of numerous genes present in the high-dimensional data sets that possibly reduce the learning algorithm's ability and the dataset's imbalanced nature, particularly with the small number of RT samples, which led to initial models failing to classify recurrent tumours accurately.Hence, the present research introduces a Balanced EOC using the Artificial Neural Network (BEOC-ANN) model to address the challenge of accurately classifying PT and RT in EOC using RNA-sequencing (RNA-Seq) data.The dataset sourced from The Cancer Genome Atlas (TCGA) database includes 374 PT samples and 5 RT samples.A pre-processing stage is implemented, using the DESeq normalization method to handle the raw HTSeq count data and filtering 112 ovarian cancer-related genes from an initial 27,620 gene features.To resolve this, the ANN model has been fine-tuned by adding dropout layers and class weights, which helped balance the dataset.The ANN model is trained with ReLU activations for input and hidden layers, sigmoid for output, and an Adam optimizer and binary cross-entropy for loss function.A significant improvement was observed when the test size increased from 10% to 30%, allowing three RT records to be recognized.The research results demonstrate that the model achieves a training accuracy of 98%, a testing accuracy of 96%, and a recall rate of 33% for RT samples.