Retinoblastoma
Conditions
Brief summary
Retinoblastoma is the most common eye cancer of childhood. Eye-preserving therapies require routine monitoring of retinoblastoma regression and recurrence to guide corresponding treatment. In the current study, we develop a deep learning algorism that can simultaneously identify retinoblastoma tumours on Retcam images and distinguish between active and inactive retinoblastoma tumours. This algorism will be validated through a prospectively collected dataset.
Detailed description
Retinoblastoma, the most common eye cancer of childhood, affects 1 in 15 000 to 1 in 18 000 live births. China has the second-largest number of patients with retinoblastoma in the world. Eye-preserving therapies have been used widely in China for approximately 15 years. Eye-preserving therapies require routine monitoring of retinoblastoma regression and recurrence to guide corresponding treatment. However, the major amount of qualified ophthalmologists are concentrated in several medical centres. Deep learning based on Retcam examination that can identify retinoblastoma will reduce screening accuracy of the local hospitals and reduce monitoring wordload. In the current study, a deep learning algorism was developed that can simultaneously identify retinoblastoma tumours on Retcam images and distinguish between active and inactive retinoblastoma tumours. This algorism will be validated through a prospectively collected dataset.
Interventions
DIAGNOSTIC_TESTDeep learning algorism
A deep learning algorism that was developed previous would be applied to identify retinoblastoma tumours on Retcam images and distinguish between active and inactive retinoblastoma tumours. The decision of two different senior ophthalmologists would be the gold standard.
Sponsors
Beijing Tongren Hospital
Study design
Observational model
COHORT
Time perspective
PROSPECTIVE
Eligibility
Sex/Gender
ALL
Age
0 Years to 5 Years
Healthy volunteers
No
Inclusion criteria
* Retinoblastoma patients undergo standard medical management.
Exclusion criteria
* The operators identified images non-assessable for a correct diagnosis, due to reasons such as blur and defocus, and excluded them from further analysis.
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Diagnosis accurcy of deep learning algorism | 1 week | The diagnosic accurcy of this deep learning algorism is the proportion of true positive and true negative in all evaluated cases |
Countries
China
Contacts
Primary ContactWenbin Wei, MD
Backup ContactRuiheng Zhang, MD
Outcome results
None listed