With over 85 million CT scans performed annually in the United States, creating tumor-related reports is a challenging and time-consuming task for radiologists. To address this need, we present RadGPT, an Anatomy-Aware Vision-Language AI Agent for generating detailed reports from CT scans. RadGPT first segments tumors, including benign cysts and malignant tumors, and their surrounding anatomical structures, then transforms this information into both structured reports and narrative reports. These reports provide tumor size, shape, location, attenuation, volume, and interactions with surrounding blood vessels and organs. Extensive evaluation on unseen hospitals shows that RadGPT can produce accurate reports, with high sensitivity/specificity for small tumor (<2 cm) detection: 80/73% for liver tumors, 92/78% for kidney tumors, and 77/77% for pancreatic tumors. For large tumors, sensitivity ranges from 89% to 97%. The results significantly surpass the state-of-the-art in abdominal CT report generation. RadGPT generated reports for 17 public datasets. Through radiologist review and refinement, we have ensured the reports' accuracy, and created the first publicly available image-text 3D medical dataset, comprising over 1.8 million text tokens and 2.7 million images from 9,262 CT scans, including 2,947 tumor scans/reports of 8,562 tumor instances. Our reports can: (1) localize tumors in eight liver sub-segments and three pancreatic sub-segments annotated per-voxel; (2) determine pancreatic tumor stage (T1-T4) in 260 reports; and (3) present individual analyses of multiple tumors--rare in human-made reports. Importantly, 948 of the reports are for early-stage tumors.

Learning a discriminative semantic space using unlabelled and noisy data remains unaddressed in a multi-label setting. We present a contrastive self-supervised learning method which is robust to data noise, grounded in the domain of variational methods. The method (VCL) utilizes variational contrastive learning with beta-divergence to learn robustly from unlabelled datasets, including uncurated and noisy datasets. We demonstrate the effectiveness of the proposed method through rigorous experiments with multi-label datasets in the face understanding domain, including one where the system is pretrained with web collected face images. Experiments include linear evaluation and fine-tuning scenarios, in addition to verification and face attribute learning tests, showing that the model learns effective embedding representations. In almost all tested scenarios, VCL surpasses the performance of state-of-the-art self-supervised methods.

We address the problem of web supervised learning, in particular for face attribute classification. Web data suffers from image set noise, due to unrelated images that may be retrieved in response to the query. We propose a semi-supervised pseudo-labeling approach where the embedding space distribution is learnt via variational contrastive learning. We use 40 Gaussian sampling heads for the 40 attributes in the CelebA dataset and apply supervised contrastive learning over a limited amount of labelled data, to address the multi-label face attribute classification problem. Soft pseudo-labeling is then used to label the unlabelled data at attribute level, followed by two-stage domain adaptation. We show that the proposed method using noisy web data brings improvements in accuracy over supervised multi-label face attribute classification in all experimental settings (over 2% points for very low-data setting). We suggest that learning the embedding distribution and the subsequent soft pseudo-labeling according to the nearest neighbors help in overcoming the noise in the unlabeled data.