OCAE and OUNET: Standard automatic optimization for medical image segmentation

Different algorithms have been developed using Deep Learning (DL) to assist clinicians. The extensive utilization of normal radiography and other modalities can significantly augment the information contained in the used dataset for training DL algorithms, thereby introducing challenges in achieving accurate diagnoses for critical diseases. Our objective is to identify an efficient and rapid model that adheres to established standards, enabling its widespread application across various pathologies and organs.

Image noise removal is a crucial preprocessing step in medical image analysis. The first stage of the proposed work involves improvements to the Convolutional Auto Encoder (CAE) that can potentially serve as a solution to the problem, among various other artificial intelligence (AI) methods. Then, the optimization of the DL model is proposed, this is a promising candidate for detecting anomalies and segmenting them accurately in medical images. In this study, we introduce a standardized approach for organ segmentation utilizing an optimized fully convolutional network (OUNET) and an Optimized CAE (OCAE). To enhance the dataset and improve image denoising, the Particle Swarm Optimization (PSO) algorithm is used, preventing information loss during the denoising process conducted by OCAE.

The segmentation results on validation data for each model clearly demonstrate the impact of the PSO algorithm, first on the CAE, and then on UNET.

The proposed improvements to CAE, optimization of DL models, and the incorporation of PSO algorithm for image denoising contribute to the creation of a novel network, OCAE + OUNET, capable of automatically segmenting various organs.