Abstract

Drugs play an indispensable role in treating liver tumors nowadays. Meanwhile, liver tumors present a significant health challenge, demanding accurate diagnostic tools that are safe for all patients, including those with iodine allergies in pharmacy or renal insufficiency. Addressing the limitations of traditional contrast-enhanced computed tomography scans, we introduce plain scan liver tumors dataset and a new model based on the unit model (YNetr model), which is named for its resemblance to a Y rotating counterclockwise. The YNetr model is the plain scan liver tumors dataset consists of multiple liver tumor plain scan segmentation datasets, meticulously assembled and annotated. Our innovation, the YNetr model, leverages wavelet transforms to extract varied frequency information, aiming to enhance diagnostic accuracy without the need for contrast agents. This model achieved a remarkable dice coefficient of 62.63 % on the plain scan liver tumors dataset, outperforming existing models by 1.22 %. Our comprehensive comparison included models like UNet 3+XNet, UNetr, and more, highlighting YNetr’s superior capability in non-contrast liver tumor segmentation. This breakthrough not only provides a safer diagnostic alternative but also improves the effectiveness of drug treatments, demonstrating the vital role of technological innovations in improving patient treatment and safety.