Multicenter Privacy-Preserving Model Training for Deep Learning Brain Metastases Autosegmentation

Objectives: This work aims to explore the impact of multicenter data heterogeneity on deep learning brain metastases (BM) autosegmentation performance, and assess the efficacy of an incremental transfer learning technique, namely learning without forgetting (LWF), to improve model generalizability without sharing raw data. Materials and methods: A total of six BM datasets from University Hospital Erlangen (UKER), University Hospital Zurich (USZ), Stanford, UCSF, NYU and BraTS Challenge 2023 on BM segmentation were used for this evaluation. First, the multicenter performance of a convolutional neural network (DeepMedic) for BM autosegmentation was established for exclusive single-center training and for training on pooled data, respectively. Subsequently bilateral collaboration was evaluated, where a UKER pretrained model is shared to another center for further training using transfer learning (TL) either with or without LWF. Results: For single-center training, average F1 scores of BM detection range from 0.625 (NYU) to 0.876 (UKER) on respective single-center test data. Mixed multicenter training notably improves F1 scores at Stanford and NYU, with negligible improvement at other centers. When the UKER pretrained model is applied to USZ, LWF achieves a higher average F1 score (0.839) than naive TL (0.570) and single-center training (0.688) on combined UKER and USZ test data. Naive TL improves sensitivity and contouring accuracy, but compromises precision. Conversely, LWF demonstrates commendable sensitivity, precision and contouring accuracy. When applied to Stanford, similar performance was observed. Conclusion: Data heterogeneity results in varying performance in BM autosegmentation, posing challenges to model generalizability. LWF is a promising approach to peer-to-peer privacy-preserving model training.

Automatized Self-Supervised Learning for Skin Lesion Screening

The incidence rates of melanoma, the deadliest form of skin cancer, have been increasing steadily worldwide, presenting a significant challenge to dermatologists. Early detection of melanoma is crucial for improving patient survival rates, but identifying suspicious lesions through ugly duckling (UD) screening, the current method used for skin cancer screening, can be challenging and often requires expertise in pigmented lesions. To address these challenges and improve patient outcomes, an artificial intelligence (AI) decision support tool was developed to assist dermatologists in identifying UD from wide-field patient images. The tool uses a state-of-the-art object detection algorithm to identify and extract all skin lesions from patient images, which are then sorted by suspiciousness using a self-supervised AI algorithm. A clinical validation study was conducted to evaluate the tool's performance, which demonstrated an average sensitivity of 93% for the top-10 AI-identified UDs on skin lesions selected by the majority of experts in pigmented skin lesions. The study also found that dermatologists confidence increased, and the average majority agreement with the top-10 AI-identified UDs improved to 100% when assisted by AI. The development of this AI decision support tool aims to address the shortage of specialists, enable at-risk patients to receive faster consultations and understand the impact of AI-assisted screening. The tool's automation can assist dermatologists in identifying suspicious lesions and provide a more objective assessment, reducing subjectivity in the screening process. The future steps for this project include expanding the dataset to include histologically confirmed melanoma cases and increasing the number of participants for clinical validation to strengthen the tool's reliability and adapt it for real-world consultation.