autoPET/CT IV Banner
Join

📋 Tasks¶

The tasks will handle lesion segmentation for whole-body hybrid imaging for a single-staging PET/CT (Task 1) or a longitudinal CT screening (Task 2). All public available data and (pre-trained) models (including foundation models) are allowed. Time points and/or imaging modality inputs can be treated individually or jointly. Participants can either develop a

  • novel method and/or
  • focus on data-centric approaches (using the provided baselines), i.e. development on pre- and post-processing pipelines.

You can participate in either one Task or in both Tasks.

🎯 Goal¶

The aim is to develop a) accurate and b) robust lesion segmentation methods that work under varying levels of human-annotated information in either single-staging whole-body PET/CT (Task 1) or a longitudinal CT screening (Task 2).



📋 Task 1: Single-staging whole-body PET/CT¶

In Task 1, we study an interactive human-in-the-loop segmentation scenario and investigate the impact of varying degrees of label information (none to multiple clicks per lesion and background) on the segmentation performance in a multi-tracer and multi-center imaging setting - similar to the previous iterations of the autoPET challenge. The specific challenge in automated segmentation of lesions in PET/CT is to avoid false-positive segmentation of anatomical structures that have physiologically high uptake while capturing all tumor lesions. This task is particularly challenging in a multitracer setting since the physiological uptake partly differs for different tracers: e.g. brain, kidney, heart for FDG and e.g. liver, kidney, spleen, submandibular for PSMA.

We will study the behaviour over 11 interactive segmentation steps. In each step, an additional standardized and pre-simulated tumor (foreground) and background click, represented as a set of 3D coordinates, will be provided alongside the input image. This process will progress incrementally from 0 clicks (1st step) to the full allocation of 10 tumor and 10 background clicks per image (11th step). Please also refer to the evaluation metrics and the structure of the input and output interfaces of the baseline containers.

Input¶
  • CT image (MHA file)
  • PET image (MHA file)
  • Lesion click(s) in foreground and background (JSON file)
Output¶
  • lesion segmentation mask (MHA file)


📋 Task 2: Longitudinal CT screening¶

In Task 2, we explore the segmentation performance to detect lesions in a follow-up scan if provided with a lesion segmentation in the baseline scan together with or without clicks of the lesions in the follow-up scan. We investigate the segmentation performance in a novel provided longitudinal CT database of >300 cases - similar to autoPET I. The specific challenge lies in the correct identification of a lesion between baseline and follow-up scan under therapy. A tumor can change shape (progression or regression), disappear (complete response) or new lesions can appear (metastasis). In addition, in some cases the differentiation between malignant and benign tissue can be difficult.

We will study the behaviour over 2 interactive segmentation steps. In the first step, only the CT images of the baseline and follow-up scan are provided together with the baseline segmentation mask (for tumor localization). For the first step no tumor click is given, i.e. empty lesion click file. In the second step, the CT images (baseline and follow-up), the baseline segmentation mask and an additional pre-simulated tumor click in the follow-up CT scan are provided. We encourage the usage of registration algorithms to align the baseline and follow-up CT scan, i.e. to transform the baseline lesion (from baseline segmentation) into the follow-up CT scan and target region (click), but this is not mandatory. The click is represented as a set of 3D coordinates on the follow-up CT image. Please also refer to the evaluation metrics and the structure of the input and output interfaces of the baseline containers.

Input¶
  • Baseline CT image (MHA file)
  • Follow-up CT image (MHA file)
  • Baseline lesion segmentation mask (MHA file)
  • Lesion click in follow-up CT image (JSON file)
Output¶
  • lesion segmentation mask of follow-up CT image (MHA file)