Directing Mutual Informative Active Learning Model to Perform Panoptic Segmentation in Lung Images

Panoptic segmentation automatically delineates particular anatomical regions of interest and facilitates numerous subsequent tasks like shape analysis and volume measurement; image segmentation is crucial to medical image analysis. In particular, the performance of segmentation algorithms has been significantly improved by the recent rapid development of deep learning techniques, which effectively use massive volumes of labelled data to build complex models. Nevertheless, a significant barrier to the application of learning-based techniques for medical images may be the challenge of acquiring manual labels for training. Researchers have looked into mutual information-based learning strategies to loosen the labelling criteria to solve this issue. Researchers outline the fundamental concepts of deep learning-based segmentation along with some of the most advanced methods available today, arranged according to the type of supervision. Given a particular segmentation objective and dataset, we aim to offer the reader some potential options for model selection, training approaches, and data modification using the Mutual informative active learning (MIALA) model for lung image segmentation. To build a better trade-off, the functionality of the proposed model is compared with other existing alternatives.