Detection of Ovarian Cancer with optimized Neural Networks using Convolutional LSTM

Abstract

Abstract: Ovarian cancer ranks as the sixth deadliest cancer in the United States, underscoring the urgent need for a non-surgical diagnostic assay with both high sensitivity and specificity. Such a breakthrough would greatly accelerate treatment and enhance the quality of life for affected individuals. In this research, an automated framework was created using optical coherence tomography (OCT) recordings to look for signs of ovarian cancer in transgenic mice. This study was carried out to aid in the remedying of this problem. In order to evaluate temporally ordered sequences of optical coherence tomography (OCT) tomograms, this advanced approach employs neural networks. Three distinct neural network-based strategies were employed, including a VGG-supported feed-forward network, a 3D convolutional neural network, and a convolutional LSTM (Long Short-Term Memory) network. Regarding the noise that is present in all OCT images, the results show that these models achieve high performance with no adjusting or feature creation needed. What a surprising discovery. The best model is a convolutional LSTM-based neural network, with a mean area under the curve (± standard error) of 0.86 ± 0.052. To our knowledge, no previous literature documents the use of machine learning to interpret depth-resolved OCT images of entire ovaries, hence this study represents an important first step in this direction. As a result, this study stands out as an innovative one. This diagnostic technique has enormous prospects for early identification and intervention in the case of this extremely deadly condition since it has the potential to be transferred from transgenic mice to human tissue. Transgenic mice are currently used as the only diagnostic tool for this disease.