Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MiceBoneChallenge: Micro-CT public dataset and six solutions for automatic growth plate detection in micro-CT mice bone scans

Nikolay Burlutskiy, Marija Kekic, Jordi de la Torre, Philipp Plewa, Mehdi Boroumand, Julia Jurkowska, Borjan Venovski, Maria Chiara Biagi, Yeman Brhane Hagos, Roksana Malinowska-Traczyk, Yibo Wang, Jacek Zalewski, Paula Sawczuk, Karlo Pintarić, Fariba Yousefi, Leif Hultin

TL;DR

Six computer vision solutions were developed that can accurately identify the location of the growth plate plane that is essential for fully automatic segmentation of trabecular bone in mice.

Abstract

Detecting and quantifying bone changes in micro-CT scans of rodents is a common task in preclinical drug development studies. However, this task is manual, time-consuming and subject to inter- and intra-observer variability. In 2024, Anonymous Company organized an internal challenge to develop models for automatic bone quantification. We prepared and annotated a high-quality dataset of 3D $μ$CT bone scans from $83$ mice. The challenge attracted over $80$ AI scientists from around the globe who formed $23$ teams. The participants were tasked with developing a solution to identify the plane where the bone growth happens, which is essential for fully automatic segmentation of trabecular bone. As a result, six computer vision solutions were developed that can accurately identify the location of the growth plate plane. The solutions achieved the mean absolute error of $1.91\pm0.87$ planes from the ground truth on the test set, an accuracy level acceptable for practical use by a radiologist. The annotated 3D scans dataset along with the six solutions and source code, is being made public, providing researchers with opportunities to develop and benchmark their own approaches. The code, trained models, and the data will be shared.

MiceBoneChallenge: Micro-CT public dataset and six solutions for automatic growth plate detection in micro-CT mice bone scans

TL;DR

Six computer vision solutions were developed that can accurately identify the location of the growth plate plane that is essential for fully automatic segmentation of trabecular bone in mice.

Abstract

Detecting and quantifying bone changes in micro-CT scans of rodents is a common task in preclinical drug development studies. However, this task is manual, time-consuming and subject to inter- and intra-observer variability. In 2024, Anonymous Company organized an internal challenge to develop models for automatic bone quantification. We prepared and annotated a high-quality dataset of 3D CT bone scans from mice. The challenge attracted over AI scientists from around the globe who formed teams. The participants were tasked with developing a solution to identify the plane where the bone growth happens, which is essential for fully automatic segmentation of trabecular bone. As a result, six computer vision solutions were developed that can accurately identify the location of the growth plate plane. The solutions achieved the mean absolute error of planes from the ground truth on the test set, an accuracy level acceptable for practical use by a radiologist. The annotated 3D scans dataset along with the six solutions and source code, is being made public, providing researchers with opportunities to develop and benchmark their own approaches. The code, trained models, and the data will be shared.

Paper Structure

This paper contains 35 sections, 1 equation, 14 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Motivation: why the challenge was set up
  3. Challenge set up
  4. Growth plate plane (GPP) definition
  5. Data: $\mu$CT scanning and 3D reconstruction
  6. Challenge logistics and evaluation
  7. Evaluation Function
  8. Challenge restrictions
  9. Submitted Solutions
  10. Data Pre-processing
  11. Modeling Approaches
  12. 3D CNN Regression Approach
  13. 2.5D Long Axis Regression
  14. 2D/2.5D Axial Plane Classification
  15. Data Augmentation
  16. ...and 20 more sections

Figures (14)

  • Figure 1: The process of bone quantification has several manual annotation steps that motivated us to organize the challenge. Growth plate plane search and then bone segmentation were the two tasks in the challenge. This paper is focused on automatic growth plate plane detection solutions.
  • Figure 2: The challenge is to predict the growth plate plane index (GPPI). The growth plate plane is defined as the lowest plane of the growth plate (the blue area) where the bone grows. In this Figure GPPI equals $175$, it is the red plane in the axial projection.
  • Figure 3: Six teams applied different pre-processing techniques, utilized 3D/2.5D/2D modeling approaches, used data augmentation, and then utilzed post processing methods including ensembling and cross validation. Finally, the performance of the teams was evaluated using a survival function for ranking the teams.
  • Figure 4: The examples of the True GPPI and the predictions by all the six teams for three bones, 5dd1c0c131, f27da128ab and 64d33d4c9c. For example, BM predicted GPPI +2 from the True GPPI for the bone 5dd1c0c131. The differences between the True GPPI and the adjacent growth plate planes are subtle and distinguishable only by experienced radiologists.
  • Figure 5: Training procedure by SN team utilizing 3D sliding window object detection approach.
  • ...and 9 more figures