Does an orthopaedic imaging question require classification, detection or segmentation?

It distinguishes classifying whether a meniscal tear exists, detecting its location with a box and segmenting precise structural boundaries.

Classification identifies image types, detection locates objects and segmentation outlines boundaries. The required reference labels and evaluation metrics depend on the question, even for the same image.
KSSTA (2022). DOI: 10.1007/s00167-021-06838-8 · Original explanatory chart based on reported data
Evidence & Original Source Image use terms

The research question

Participants and Methods

This editorial explains three deep-learning tasks: image classification, object location/type detection and pixel-level segmentation. Published examples illustrate their differences. The article explains imaging-AI design principles without a new patient cohort.

Main Findings

It distinguishes classifying whether a meniscal tear exists, detecting its location with a box and segmenting precise structural boundaries. Tasks differ in reference-labelling methods and annotation workload. A combined approach first segments a structure and then classifies the region.

Implications for clinical and surgical practice

Classification can support OA grading, detection surgical-tool localisation, and segmentation joint or muscle measurements. A clear research question is needed to define reference data and performance measures.

Limitations

The article does not develop a new independent model or compare clinical effects. Performance from example papers cannot be interpreted as pooled performance of this editorial.

Let's put it briefly.

Which clinical question does this address?

Keywords

Deep Learning
Orthopedic Procedures
Orthopedics
Artificial Intelligence
Machine Learning
AI
classification
object detection

Abstract · Original text

Sunho Ko, Ayoosh Pareek, Du Hyun Ro, Yining Lu, Christopher L Camp, R Kyle Martin, Aaron J Krych. Artificial intelligence in orthopedics: three strategies for deep learning with orthopedic specific imaging. Knee surgery, sports traumatology, arthroscopy : official journal of the ESSKA. 2022;30(3):758-761. doi:10.1007/s00167-021-06838-8.

No abstract is listed for this article.

Read abstract on PubMed

The original paper.

Artificial intelligence in orthopedics: three strategies for deep learning with orthopedic specific imaging

Sunho Ko, Ayoosh Pareek, Du Hyun Ro, Yining Lu, Christopher L. Camp, R. Kyle Martin, Aaron J. Krych
KSSTA · 2022

DOI · 10.1007/s00167-021-06838-8 PubMed Based data

Academic Discussion

Research question and design