TwinLiverNet: Predicting TACE Treatment Outcome from CT scans for Hepatocellular Carcinoma using Deep Capsule...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Predicting treatment response is crucial for liver cancer (hepatocellular carcinoma, HCC) therapy planning.
• Trans-arterial chemoembolization (TACE) is a primary treatment for unresectable HCC.
• Radiological assessment of TACE response via CT scans is subjective and time-consuming.

Purpose of the Study:

• To develop a deep neural network, TwinLiverNet, for automated prediction of TACE treatment outcomes in HCC.
• To leverage multi-phase contrast-enhanced CT scans for improved treatment response prediction.
• To reduce inter-operator variability in assessing treatment response.

Main Methods:

• TwinLiverNet integrates 3D convolutions and capsule networks.
• The model processes late arterial and delayed phases of contrast-enhanced CT scans simultaneously.
• Experiments were conducted on a dataset of 126 HCC lesions.

Main Results:

• TwinLiverNet achieved an average accuracy of 82% in predicting complete response to TACE.
• Utilizing multiple CT phases (late arterial and delayed) improved performance by over 12 percentage points.
• Capsule layers in the model mitigated overfitting and enhanced predictive accuracy.

Conclusions:

• TwinLiverNet offers a reliable and automated method for predicting TACE treatment response in HCC.
• The integration of multi-phase CT data significantly boosts prediction performance.
• This AI tool can support radiologists, enhancing the accuracy and consistency of treatment outcome assessment.