Susana Lai-Yuen

Accurate detection of breast cancer is essential, as it remains a leading cause of cancer-related mortality worldwide. Ultrasound is adopted due to its accessibility, safety, and capability, complementing mammography and MRI. Yet interpretation is challenging because images are operator-dependent. This has motivated deep learning (DL) to improve lesion detection, classification, and segmentation. Although DL achieves high accuracy under noise-free conditions, real-world reliability remains underexplored because acquisition factors and speckle noise degrade images, increasing missed-malignancy risk. To address the limited reliability of breast ultrasound classifiers under noisy acquisition conditions, we propose a noise-aware training framework that evaluates model robustness under controlled Gaussian, Poisson, and speckle noise to simulate clinically relevant degradations. We used the publicly available breast ultrasound images (BUSI) dataset, which contains 780 ultrasound images collected in 2018 from 600 female patients, and evaluated a custom convolutional neural network (CNN) and a pretrained Inception V3 model. The framework combines controlled noise modeling, robustness assessment, and noise-matched training across different noise types and severities. Performance was assessed via accuracy and malignant recall. Results show that Gaussian noise most affected the performance of the custom CNN, reducing accuracy by 36.1% at 5 dB signal-to-noise ratio (SNR), while Poisson noise most affected Inception V3, reducing accuracy by 35.2%. Speckle noise produced smaller accuracy changes, but malignant recall declined under high noise. Noise-matched training improved accuracy by up to 56.5% at 5 dB, though gains were smaller at reduced noise. These findings show that noise-aware, recall-preserving evaluation is necessary for safer breast ultrasound analysis.

Deep learning methods have become very successful at solving many complex tasks such as image classification and segmentation, speech recognition and machine translation. Nevertheless, manually designing a neural network for a specific problem is very difficult and time-consuming due to the massive hyperparameter search space, long training times, and lack of technical guidelines for the hyperparameter selection. Moreover, most networks are highly complex, task specific and over-parametrized. Recently, multiobjective neural architecture search (NAS) methods have been proposed to automate the design of accurate and efficient architectures. However, they only optimize either the macro- or micro-structure of the architecture requiring the unset hyperparameters to be manually defined, and do not use the information produced during the optimization process to increase the efficiency of the search. In this work, we propose EMONAS, an Efficient MultiObjective Neural Architecture Search framework for the automatic design of neural architectures while optimizing the network's accuracy and size. EMONAS is composed of a search space that considers both the macro- and micro-structure of the architecture, and a surrogate-assisted multiobjective evolutionary based algorithm that efficiently searches for the best hyperparameters using a Random Forest surrogate and guiding selection probabilities. EMONAS is evaluated on the task of 3D cardiac segmentation from the MICCAI ACDC challenge, which is crucial for disease diagnosis, risk evaluation, and therapy decision. The architecture found with EMONAS is ranked within the top 10 submissions of the challenge in all evaluation metrics, performing better or comparable to other approaches while reducing the search time by more than 50% and having considerably fewer number of parameters.