| Summary: | In the domain of medical diagnostics, medical images serve as invaluable tools for elucidating various pathological conditions. However, the process of image segmentation, which involves partitioning these images into distinct regions of interest, is profoundly challenging due to the intricate and diverse pixel features inherent to medical imagery. While Convolutional Neural Networks (CNNs) have exhibited proficiency in this task, they come at a significant cost in terms of computational resources and the requisite sizeable datasets for training, rendering them computationally demanding. To mitigate these impediments, this study introduces an innovative paradigm rooted in Object-oriented Programming (OOP) principles, denoted as the Object-oriented Pixel Descriptor (OOPD). This methodology augments the information encapsulated within medical images, synergizing with a deep neural network architecture to yield the Object-oriented Deep Neural Network (OODNN) segmentation model. In a rigorous evaluation encompassing three distinct medical image datasets, each characterized by a limited dataset size, OODNN demonstrates substantial prowess, achieving a notable balanced accuracy of 0.771. This performance is analogous to contemporary state-of-the-art CNN-based models, albeit with a marked reduction in training duration and computational demands. These findings present a promising avenue for the advancement of medical image segmentation, with potential ramifications for augmenting the diagnostic capabilities of healthcare practitioners. © 2023 IEEE.
|
|---|