Fully Automatic Liver and Tumor Segmentation from CT Image Using an AIM-Unet.

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Artificial Intelligence in Medicine

Background:

• Liver segmentation in computed tomography (CT) images is challenging due to variations in shape, border, and density.
• Accurate segmentation is crucial for diagnosing liver conditions and planning treatments.

Purpose of the Study:

• To propose the Adding Inception Module-Unet (AIM-Unet) model for automated liver and liver tumor segmentation in abdominal CT scans.
• To evaluate the performance of the AIM-Unet model against specialist-marked segmentations using established metrics.

Main Methods:

• Developed a hybrid deep learning model, AIM-Unet, integrating Unet and Inception architectures.
• Trained and tested the model on four diverse liver CT datasets, including open-source (CHAOS, LIST, 3DIRCADb) and a custom dataset.
• Quantitatively assessed segmentation accuracy using Dice Similarity Coefficient (DSC), Jaccard Similarity Coefficient (JSC), and Accuracy (ACC).

Main Results:

• The AIM-Unet model achieved high liver segmentation performance on the CHAOS dataset, with DSC of 97.86%, JSC of 96.10%, and ACC of 99.75%.
• Tumor segmentation yielded DSC metrics of 75.6% on the LiST dataset and 65.5% on the 3DIRCADb dataset.
• Performance comparison with previous studies demonstrated the effectiveness of the proposed AIM-Unet model.

Conclusions:

• The AIM-Unet model shows significant potential as an auxiliary tool for physicians in liver segmentation and liver tumor detection from CT scans.
• The developed model's architecture is adaptable for segmentation tasks in other organs and medical imaging fields.
• This research contributes to advancing computer-assisted diagnosis in medical imaging.