MTSAIL & LEAF     

Dynamic and longitudinal data acquisition and image analysis have seen significant growth in healthcare and related fields, and have been applied to a range of applications, including real-time imaging, motion analysis, and the analysis of follow-up lesion tracking. Recent advances in imaging and data acquisition have led to an increase in temporal modeling. Time-series data, which is collected through continuous real-time monitoring or long-term follow-up of a subject, requires specialized methods for extracting and connecting related datasets through the analysis of temporal information. These methods are essential for effectively studying time-series data and gaining insights from it.

For short- or real-time scales video and motion data tracking/processing (MTSAIL), the proliferation of dynamic data acquisition frameworks often produces high-dimensional datasets. This trend underscores the need for effective methods for analyzing and interpreting time-series data. The analysis of time-series data also presents a number of algorithmic challenges, including insufficient data structure, irregular sampling, inaccurate motion tracking, spatiotemporal misalignment, and multimodality data synthesis. These issues can make it difficult to extract meaningful insights from time-series datasets and highlight the need for robust methods for addressing them.

For longer-time scales for lesion tracking (LEAF), the measurement of structures that are suspicious for malignancy (lesions, tumors, lymph nodes, etc.) in longitudinal radiologic imaging studies (e.g., CT/MR/PET/Ultrasound) are crucial for charting the course of therapy in patients with cancer. Unfortunately, the manual assessment is significantly burdensome for radiologists in terms of both time and effort. Furthermore, linear measurements of lesion size are often subject to significant inter-reader variability; volumetric measurements have lower variabilities, but they are more sensitive to interval change, and cumbersome to manually accomplish during a busy clinical day. With recent developments in AI, many challenges remain, including universal lesion/abnormality detection across multiple organs or anatomical structures, identification and detection of other structures related to cancer staging (e.g., lymph nodes), effective incorporation of longitudinal information from multiple studies, the fusion of information from multi-modal imaging and reporting data, and the performance robustness to heterogeneous data (across patients, scanners, institutional imaging protocols, contrast phases, etc).

This workshop aims to identify and showcase the latest advances in temporal data processing and longitudinal suspicious structures analysis algorithms, new datasets, as well as application areas. By collecting new research and discussing the challenges and potential problems facing time-series data, the workshop aims to help shape the direction of future research in this field.