About this Research Topic
Sometimes, volumetric measurements are needed for treatment planning; this can also be obtained from appropriate image segmentation. There is another imaging modality that is quite popular among the clinicians because of no radiation and less cost, that is Ultrasound (US). Despite the benefits of US imaging, clinicians and surgeons face several challenges in using US images.
Some of them are:
1) The quality of US image is generally poor due to the widespread presence of speckle noise and shadows. Specifically, the shadow of the ribs and lungs may cover certain portions of the liver, adding difficulty to the analysis.
2) Anatomical boundaries are generally not clear due to the frequent appearance of imaging artifacts and poorly reflected sound waves from tissues of varying echogenicity. For instance, the border of the liver often overlaps with the diaphragm making the precise edge of the liver obscure.
3) The shape and size of the regions of interest (ROI) may vary significantly depending on the orientation and location of the US probe.
4) The presence of other anatomical structures with similar echogenicity in the vicinity of ROI makes it difficult to be delineated, for instance, the kidney, the spleen, etc.
5) It is difficult to identify small ROI (e.g., hepatic lesions, portal vein branches, hepatic artery) due to the low contrast nature of US.
All these challenges make the manual analysis of real-time liver US difficult, time-consuming and operator dependent. One way to overcome these hurdles is by precisely delineating the ROI (i.e., liver) in real-time, allowing the clinician to detect the liver amongst other organs/tissues on the US machine display. This outlining could also help them perform a straight-forward analysis and diagnosis of hepatic conditions. Surgeries and therapeutic procedures could further use the real-time US segmentation methods to increase the accuracy of the procedure while reducing the damage of healthy tissues.
In general, manual segmentation is quite tedious and hectic, thus, computer generated segmentation algorithms are needed. Over the years, several conventional segmentation algorithms based on different approaches have been proposed that include region growing, thresholding, watershed, active contours, clustering, etc.
Recently, several deep learning (DL)- based techniques have drawn wider attention in automating the process and achieving higher segmentation accuracy. These DL-methods would certainly augment artificial intelligence to take a leap forward in the direction of automating the segmentation process.
Since the field of medical image segmentation is quite broad, we aim to focus on the following topics, both clinical and technical:
• Hepatocellular carcinoma treatment planning for intervention, resection and transplant. We would expect contributions on present treatments for HCC that include: planning, potential treatments, shortcomings/challenges, mortality or morbidity statistics.
• Memory efficient 3D convolutional neural networks for liver CT, MR and US image segmentation We would expect contributions on light weight and memory efficient neural networks. Additionally, we would expect contributions on real-time liver segmentation methods that would really be considered in artificial intelligence.
• Uncertainty estimation in neural networks. There have been several networks; we would expect a couple of review papers mentioning the limitations of the present neural networks. Additionally, we could expect original articles potentially degerming to estimate the uncertainty in a neural network.
• Uncertainty estimation in segmentation. We would expect contributions about the possible uncertainties of the neural networks while applied for real-time liver segmentation. This could include performance (accuracy), user-friendliness, etc.
Keywords: Liver, hepatocellular carcinoma, prognosis, treatment planning, artificial intelligence, real-time, image segmentation, neural network, deep learning
Important Note: All contributions to this Research Topic must be within the scope of the section and journal to which they are submitted, as defined in their mission statements. Frontiers reserves the right to guide an out-of-scope manuscript to a more suitable section or journal at any stage of peer review.