A study of generalization and compatibility performance of 3D U-Net segmentation on multiple heterogeneous liver CT datasets

Background Most existing algorithms have been focused on the segmentation from several public Liver CT datasets scanned regularly (no pneumoperitoneum and horizontal supine position). This study primarily segmented datasets with unconventional liver shapes and intensities deduced by contrast phases, irregular scanning conditions, different scanning objects of pigs and patients with large pathological tumors, which formed the multiple heterogeneity of datasets used in this study. Methods The multiple heterogeneous datasets used in this paper includes: (1) One public contrast-enhanced CT dataset and one public non-contrast CT dataset; (2) A contrast-enhanced dataset that has abnormal liver shape with very long left liver lobes and large-sized liver tumors with abnormal presets deduced by microvascular invasion; (3) One artificial pneumoperitoneum dataset under the pneumoperitoneum and three scanning profiles (horizontal/left/right recumbent position); (4) Two porcine datasets of Bama type and domestic type that contains pneumoperitoneum cases but with large anatomy discrepancy with humans. The study aimed to investigate the segmentation performances of 3D U-Net in: (1) generalization ability between multiple heterogeneous datasets by cross-testing experiments; (2) the compatibility when hybrid training all datasets in different sampling and encoder layer sharing schema. We further investigated the compatibility of encoder level by setting separate level for each dataset (i.e., dataset-wise convolutions) while sharing the decoder. Results Model trained on different datasets has different segmentation performance. The prediction accuracy between LiTS dataset and Zhujiang dataset was about 0.955 and 0.958 which shows their good generalization ability due to that they were all contrast-enhanced clinical patient datasets scanned regularly. For the datasets scanned under pneumoperitoneum, their corresponding datasets scanned without pneumoperitoneum showed good generalization ability. Dataset-wise convolution module in high-level can improve the dataset unbalance problem. The experimental results will facilitate researchers making solutions when segmenting those special datasets. Conclusions (1) Regularly scanned datasets is well generalized to irregularly ones. (2) The hybrid training is beneficial but the dataset imbalance problem always exits due to the multi-domain homogeneity. The higher levels encoded more domain specific information than lower levels and thus were less compatible in terms of our datasets.

Efficient knowledge distillation for liver CT segmentation using growing assistant network

Segmentation has been widely used in diagnosis, lesion detection, and surgery planning. Although deep learning (DL)-based segmentation methods currently outperform traditional methods, most DL-based segmentation models are computationally expensive and memory inefficient, which are not suitable for the intervention of liver surgery. To address this issue, a simple solution is to make a segmentation model very small for the fast inference time, however, there is a trade-off between the model size and performance. In this paper, we propose a DL-based real- time 3-D liver CT segmentation method, where knowledge distillation (KD) method, referred to as knowledge transfer from teacher to student models, is incorporated to compress the model while preserving the performance. Because it is known that the knowledge transfer is inefficient when the disparity of teacher and student model sizes is large, we propose a growing teacher assistant network (GTAN) to gradually learn the knowledge without extra computational cost, which can efficiently transfer knowledges even with the large gap of teacher and student model sizes. In our results, dice similarity coefficient of the student model with KD improved 1.2% (85.9% to 87.1%) compared to the student model without KD, which is a similar performance of the teacher model using only 8% (100k) parameters. Furthermore, with a student model of 2% (30k) parameters, the proposed model using the GTAN improved the dice coefficient about 2% compared to the student model without KD, with the inference time of 13ms per case. Therefore, the proposed method has a great potential for intervention in liver surgery, which also can be utilized in many real-time applications.