Assessment of a novel computer software in diagnosing radiocarpal osteoarthritis on plain radiographs of patients with previous distal radius fracture
(2020) In Osteoarthritis and Cartilage Open 2(4).- Abstract
- Objective: Osteoarthritis (OA) has primarily been diagnosed with plain radiographs assessed visually by examiners with regard to joint space width and presence of subchondral sclerosis, cysts and osteophytes. The increasing use of artificial intelligence has seen software developed to examine plain radiographs for diagnosing OA, based on observed OA-associated subchondral bone microarchitecture changes. A software for computerized texture analysis has been developed to identify knee OA. The aim of this study was to assess the software's ability to identify radiocarpal OA.
Design: Presence of radiocarpal OA on 63 wrist radiographs of patients with a previous distal radius fracture was assessed independently by two surgeons... (More) - Objective: Osteoarthritis (OA) has primarily been diagnosed with plain radiographs assessed visually by examiners with regard to joint space width and presence of subchondral sclerosis, cysts and osteophytes. The increasing use of artificial intelligence has seen software developed to examine plain radiographs for diagnosing OA, based on observed OA-associated subchondral bone microarchitecture changes. A software for computerized texture analysis has been developed to identify knee OA. The aim of this study was to assess the software's ability to identify radiocarpal OA.
Design: Presence of radiocarpal OA on 63 wrist radiographs of patients with a previous distal radius fracture was assessed independently by two surgeons experienced in assessing radiographs, and classified according to Kellgren-Lawrence (38 OA, 25 no OA). First, the computer software, not previously trained to identify wrist OA, assessed presence of radiocarpal OA on the 63 radiographs. In a second step, 144 labeled wrist radiographs with and without radiocarpal OA was used to train the computer software. Presence of OA on the original 63 radiographs were then reassessed by the software. Sensitivity, specificity and area under the curve (AUC) were calculated to determine the software's ability to discriminate between cases with and without OA.
Results: Before training, sensitivity was 76% (95% CI 59–88), specificity 25% (10–47), and AUC 0.50 (0.35–0.65). After training, sensitivity was 46% (29–63), specificity 70% (47–87), and AUC 0.58 (0.43–0.73).
Conclusion: The software for computerized texture analysis of subchondral bone developed to identify knee OA could not detect OA of the radiocarpal joint. (Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/record/e701c2ec-31db-4aef-b0b0-d605bd35f142
- author
- Ali, Muhanned LU ; Brogren, Elisabeth LU and Atroshi, Isam LU
- organization
- publishing date
- 2020
- type
- Contribution to journal
- publication status
- published
- subject
- keywords
- Osteoarthritis, artificial intelligence, distal radius fracture
- in
- Osteoarthritis and Cartilage Open
- volume
- 2
- issue
- 4
- pages
- 4 pages
- publisher
- Elsevier
- external identifiers
-
- scopus:85108525754
- ISSN
- 2665-9131
- DOI
- 10.1016/j.ocarto.2020.100112
- language
- English
- LU publication?
- yes
- id
- e701c2ec-31db-4aef-b0b0-d605bd35f142
- date added to LUP
- 2022-11-15 21:41:54
- date last changed
- 2023-09-19 04:00:06
@article{e701c2ec-31db-4aef-b0b0-d605bd35f142, abstract = {{Objective: Osteoarthritis (OA) has primarily been diagnosed with plain radiographs assessed visually by examiners with regard to joint space width and presence of subchondral sclerosis, cysts and osteophytes. The increasing use of artificial intelligence has seen software developed to examine plain radiographs for diagnosing OA, based on observed OA-associated subchondral bone microarchitecture changes. A software for computerized texture analysis has been developed to identify knee OA. The aim of this study was to assess the software's ability to identify radiocarpal OA.<br/><br/>Design: Presence of radiocarpal OA on 63 wrist radiographs of patients with a previous distal radius fracture was assessed independently by two surgeons experienced in assessing radiographs, and classified according to Kellgren-Lawrence (38 OA, 25 no OA). First, the computer software, not previously trained to identify wrist OA, assessed presence of radiocarpal OA on the 63 radiographs. In a second step, 144 labeled wrist radiographs with and without radiocarpal OA was used to train the computer software. Presence of OA on the original 63 radiographs were then reassessed by the software. Sensitivity, specificity and area under the curve (AUC) were calculated to determine the software's ability to discriminate between cases with and without OA.<br/><br/>Results: Before training, sensitivity was 76% (95% CI 59–88), specificity 25% (10–47), and AUC 0.50 (0.35–0.65). After training, sensitivity was 46% (29–63), specificity 70% (47–87), and AUC 0.58 (0.43–0.73).<br/><br/>Conclusion: The software for computerized texture analysis of subchondral bone developed to identify knee OA could not detect OA of the radiocarpal joint.}}, author = {{Ali, Muhanned and Brogren, Elisabeth and Atroshi, Isam}}, issn = {{2665-9131}}, keywords = {{Osteoarthritis; artificial intelligence; distal radius fracture}}, language = {{eng}}, number = {{4}}, publisher = {{Elsevier}}, series = {{Osteoarthritis and Cartilage Open}}, title = {{Assessment of a novel computer software in diagnosing radiocarpal osteoarthritis on plain radiographs of patients with previous distal radius fracture}}, url = {{http://dx.doi.org/10.1016/j.ocarto.2020.100112}}, doi = {{10.1016/j.ocarto.2020.100112}}, volume = {{2}}, year = {{2020}}, }