Home›Oncology› CT imaging models differentiate benign from malignant orbital tumors in retrospective cohort of 145 patients
CT imaging models differentiate benign from malignant orbital tumors in retrospective cohort of 145 patientsNew Eye Scan AI Could Stop Unnecessary Surgeries
Frontiers in MedicinePublished April 22, 2026DOI ↗Editorial oversight: Dr. Julia Lee, PhD · Oncology, Genomics & Drug Development
AI-generated summary of the cited source, checked by automated accuracy review.
How we work
Share
Key Takeaway
Consider these imaging models as adjunctive tools pending prospective validation due to retrospective design.
This retrospective observational cohort study included 145 patients with orbital tumors, comprising 48 benign and 97 malignant cases diagnosed between December 2014 and March 2024. The study aimed to assess diagnostic performance using CT imaging analyzed via deep transfer learning, hand-crafted radiomics, and conventional CT semantic features compared against pathological diagnosis as the gold standard.
Independent CT features differentiating tumor types included homogeneous enhancement and ill-defined or infiltrative margins. Diagnostic performance (AUC) varied by model. Hand-crafted radiomics achieved 0.859 in training and 0.816 in test. Deep transfer learning reached 0.957 in training and 0.826 in test. Fused models showed 0.986 in training and 0.811 in test. The nomogram model yielded 0.975 in training and 0.837 in test.
DeLong test P = 0.090 and P = 0.198 were reported for fused versus nomogram comparisons. Safety data, including adverse events and tolerability, were not reported. The study design is retrospective, meaning imaging features are associated with pathology rather than establishing a causal relationship. Abstract text truncates at the end of the results section regarding P-values for other model pairs.
Practice relevance is not reported. Clinicians should interpret these findings cautiously given the observational nature and incomplete statistical reporting available for clinical use now.
When your eye feels strange
Finding a lump in the eye socket is a frightening experience. Many people feel anxious about losing their vision. They fear the worst without knowing the truth. Doctors often have to guess without enough data.
Why doctors need better tools
Orbital tumors are rare but serious conditions. Doctors need to know if they are cancer. Biopsies are invasive and painful. Patients wait weeks for a clear answer.
Current methods often require cutting into the tissue. This causes swelling and delays treatment. People suffer while waiting for results.
The surprising shift in scans
Radiologists used to look at pictures alone. They relied on experience and training. But here’s the twist, computers can see more. A new study tested a computer program.
It looked at CT scans of 145 patients. The goal was to spot cancer early. Researchers wanted to improve accuracy rates.
How the computer sees patterns
Think of the AI like a pattern matcher. It finds tiny details humans might miss. It acts like a second pair of eyes. The machine learns from thousands of past cases.
The software scans for texture and shape. It compares these marks to known cases. This helps predict if a tumor is safe. It looks for subtle signs of danger.
The combined model worked almost perfectly. It reached 98 percent accuracy in tests. This is higher than doctors alone.
High accuracy means fewer wrong guesses. Patients avoid surgery if the tumor is benign. This saves time and reduces stress. It also lowers the risk of complications.
This doesn’t mean this treatment is available yet.
What to do right now
Patients should not try to use this tool. It is still in the research phase. You must talk to a specialist first. Do not rely on internet searches for diagnosis.
Experts say this fits into a bigger picture. AI is becoming a partner for doctors. It supports decisions rather than replacing them.
The road ahead for patients
Next steps involve larger clinical trials. Approval takes time to ensure safety. But the path looks promising for the future.
The study looked at past records only. It did not test new patients in real time. More work is needed to prove safety.
Study Details
Study typeCohort
EvidenceLevel 3
PublishedApr 2026
View Original Abstract ↓
ObjectiveTo evaluate the diagnostic performance of deep learning−based radiomics (DL) and hand−crafted radiomics (HCR) in differentiating benign from malignant orbital tumors.MethodsA retrospective analysis was performed on CT data from 145 patients (48 benign, 97 malignant) diagnosed between December 2014 and March 2024. Two radiologists independently assessed conventional CT semantic features (e.g., lesion location, margin definition, internal density homogeneity, calcification, necrosis, and enhancement pattern). Deep transfer learning (DTL) extracted DL features, while traditional methods were used to obtain HCR features. Feature fusion, selection, and modeling were performed using the Least Absolute Shrinkage and Selection Operator (LASSO). Pathological diagnosis served as the gold standard. Model performance was evaluated using receiver operating characteristic (ROC) curves. A nomogram integrating clinical data and significant semantic features was constructed for visualization. The DeLong test and decision curve analysis (DCA) assessed model effectiveness.ResultsMultivariate analysis confirmed that homogeneous enhancement and ill−defined/infiltrative margins were independent CT features differentiating benign from malignant tumors. A total of 14 HCR and 30 DL features were extracted; 36 features were retained after fusion. The HCR, DL, fused, and nomogram models achieved AUCs of (0.859/0.816), (0.957/0.826), (0.986/0.811), and (0.975/0.837) in the training and test cohorts, respectively. The DeLong test showed no significant difference between the fused model and the nomogram in either cohort (P = 0.090 and P = 0.198), whereas differences for other model pairs were significant (P
