How to overcome the data limited segmentation in abdominal CT: Multi-planar UNet coupled with augmented contrast-boosting

The quantity and variety of CT imaging data are essential components for effective AI-model training. However, the availability of high-quality CT images for organ segmentation is quite constrained, and the AI-based organ segmentation could be impacted by the varying intensity of contrast agents. Therefore, to improve the robustness of the segmentation both with and without a contrast agent, as well as to solve the data shortage issue, we proposed a multi-planar UNet with augmented contrast-boosting technique. A contrast-enhanced CT dataset was utilized to develop the AI-model, while a non-contrast CT dataset was used to assess the model's performance. The results of axial-plane UNet and multi-planar UNet were evaluated quantitatively using a dice-similarity coefficient. Furthermore, the effect of augmented contrast-boosting on the AI-model performance was analyzed. Throughout the contrast-boosting algorithm, even though only 30 patients' CT data were used, diversity and the amount of data were both increased, and it improved the mean dice-value of axial-plane UNet by 20% in non-contrast CT data. The mean dice-value was over 90% even in non-contrast CT data when multi-planar UNet was used, and it was 5% higher than when the augmented contrast-boosting wasn't applied. Our findings demonstrate that the spleen could be precisely segmented using multi-planar UNet coupled with augmented contrast-boosting, even with a small amount of CT dataset. The results would positively impact conventional AI-based segmentation strategy and its robustness. Any program employing the proposed method may see greater benefits from reducing the burden of large-scale dataset preparation, improving the AI-model training efficiency.