Subtyping breast lesions via collective intelligence based long-tailed recognition in ultrasound

Breast lesions display a wide spectrum of histological subtypes. Recognizing these subtypes is vital for optimizing patient care and facilitating tailored treatment strategies compared to a simplistic binary classification of malignancy. However, this task relies on invasive biopsy tests, which carry inherent risks and can lead to over-diagnosis, unnecessary expenses, and pain for patients. To avoid this, we propose to infer lesion subtypes from ultrasound images directly. Meanwhile, the incidence rates of different subtypes exhibit a skewed long-tailed distribution that presents substantial challenges for effective recognition. Inspired by collective intelligence in clinical diagnosis to handle complex or rare cases, we proposed a framework–CoDE–to amalgamate diverse expertise of different backbones to bolster robustness across varying scenarios for automated lesion subtyping. It utilizes dual-level balanced individual supervision to fully exploit prior knowledge while considering class imbalance. It is also equipped with a batch-based online competitive distillation module to stimulate dynamic knowledge exchange. Experimental results demonstrate that the model surpassed the state-of-the-art approaches by more than 7.22% in F1-score facing a challenging breast dataset with an imbalance ratio as high as 47.9:1.