Adaptive Evolutionary Optimization of Deep Learning Architectures for Focused Liver Ultrasound Image Segmentation
Ali Zifan,
Katelyn Zhao,
Madilyn Lee,
Zihan Peng,
Laura J. Roney,
Sarayu Pai,
Jake T. Weeks,
Michael S. Middleton,
Ahmed El Kaffas,
Jeffrey B. Schwimmer,
Claude B. Sirlin
Affiliations
Ali Zifan
Division of Gastroenterology and Hepatology, University of California San Diego, San Diego, CA 92093, USA
Katelyn Zhao
Division of Gastroenterology and Hepatology, University of California San Diego, San Diego, CA 92093, USA
Madilyn Lee
Division of Gastroenterology and Hepatology, University of California San Diego, San Diego, CA 92093, USA
Zihan Peng
Division of Gastroenterology and Hepatology, University of California San Diego, San Diego, CA 92093, USA
Laura J. Roney
Division of Gastroenterology and Hepatology, University of California San Diego, San Diego, CA 92093, USA
Sarayu Pai
Division of Gastroenterology and Hepatology, University of California San Diego, San Diego, CA 92093, USA
Jake T. Weeks
Liver Imaging Group, Department of Radiology, University of California San Diego, San Diego, CA 92093, USA
Michael S. Middleton
Liver Imaging Group, Department of Radiology, University of California San Diego, San Diego, CA 92093, USA
Ahmed El Kaffas
Liver Imaging Group, Department of Radiology, University of California San Diego, San Diego, CA 92093, USA
Jeffrey B. Schwimmer
Department of Pediatrics, Division of Gastroenterology, Hepatology, and Nutrition, University of California San Diego School of Medicine, La Jolla, CA 92093, USA
Claude B. Sirlin
Liver Imaging Group, Department of Radiology, University of California San Diego, San Diego, CA 92093, USA
Background: Liver ultrasound segmentation is challenging due to low image quality and variability. While deep learning (DL) models have been widely applied for medical segmentation, generic pre-configured models may not meet the specific requirements for targeted areas in liver ultrasound. Quantitative ultrasound (QUS) is emerging as a promising tool for liver fat measurement; however, accurately segmenting regions of interest within liver ultrasound images remains a challenge. Methods: We introduce a generalizable framework using an adaptive evolutionary genetic algorithm to optimize deep learning models, specifically U-Net, for focused liver segmentation. The algorithm simultaneously adjusts the depth (number of layers) and width (neurons per layer) of the network, dropout, and skip connections. Various architecture configurations are evaluated based on segmentation performance to find the optimal model for liver ultrasound images. Results: The model with a depth of 4 and filter sizes of [16, 64, 128, 256] achieved the highest mean adjusted Dice score of 0.921, outperforming the other configurations, using three-fold cross-validation with early stoppage. Conclusions: Adaptive evolutionary optimization enhances the deep learning architecture for liver ultrasound segmentation. Future work may extend this optimization to other imaging modalities and deep learning architectures.
