Segmentation of kidney ablation zone using deep learning in CT images

Kidney tumor thermal ablation procedures create an ablation zone that is planned to cover the tumor and destroy malignant cells. Incorrect identification of the margins of this zone can lead to incomplete treatment, increasing the risk of recurrence, or excessive damage to healthy tissue, leading to complications such as renal dysfunction. Thus, accurate segmentation of the Kidney Ablation Zone (KAZ) is vital for assessing the treatment's efficacy and planning further interventions if necessary. Despite the significant importance of this issue, no research has been conducted yet on the segmentation of KAZ. This research proposes an advanced deep learning-based approach utilizing the Attention U-Net architecture to segment KAZ and address this problem. The proposed workflow leverages the strengths of the U-Net architecture, improved with attention mechanisms, to enhance the network’s ability to focus on the most relevant regions of the images, thereby achieving appropriate segmentations. Our model was trained and evaluated on a local dataset from the London Health Sciences Centre (London, Canada) and comprised 76 patients' annotated ablation zones in kidney CT images. Quantitative analysis demonstrated that the Attention U-Net achieved promising performance metrics, including a 0.7 Dice similarity coefficient and a mean absolute boundary distance of 0.97 mm indicating its robustness and reliability in clinical settings. Furthermore, qualitative results showed that our approach effectively delineates ablation zones, providing clear and accurate boundaries critical for post-procedural assessment and planning.