Background: There are no reliable predictive models for recurrence after transsphenoidal surgery (TSS) for Cushing’s disease (CD). Objectives: This study aimed to develop machine learning (ML)-based predictive models for CD recurrence after initial TSS and to evaluate their performance. Method: A total of 354 CD patients were included in this retrospective, supervised learning, data mining study. Predictive models for recurrence were developed according to 17 variables using 7 algorithms. Models were evaluated based on the area under the receiver operating characteristic curve (AUC). Results: All patients were followed up for over 12 months (mean ± SD 43.80 ± 35.61). The recurrence rate was 13.0%. Age (p < 0.001), postoperative morning serum cortisol nadir (p = 0.002), and postoperative (p < 0.001) and preoperative (p = 0.04) morning adrenocorticotropin (ACTH) level were significantly related to recurrence. AUCs of the 7 models ranged from 0.608 to 0.781. The best performance (AUC = 0.781, 95% CI 0.706, 0.856) appeared when 8 variables were introduced to the random forest (RF) algorithm, which was much better than that of logistic regression (AUC = 0.684, p = 0.008) and that of using only postoperative morning serum cortisol (AUC = 0.635, p < 0.001). According to the feature selection algorithms, the top 3 predictors were age, postoperative serum cortisol, and postoperative ACTH. Conclusions: Using ML-based models for prediction of the recurrence after initial TSS for CD is feasible, and RF performs best. The performance of most of ML-based models was significantly better than that of some conventional models.

Context: The resection plan of pituitary adenoma (PA) needs preoperative observation of the sellar region. Radiomics prediction requires high-quality segmentations. Manual delineation is time-consuming and subject to rater variability. Objective: This work aims to create an automated segmentation method for the sellar region, several tools to extract invasiveness-related features, and evaluate their clinical usefulness by predicting the tumor consistency. Methods: Patients included were diagnosed with pituitary adenoma at Peking Union Medical College Hospital. A deep convolutional neural network, called gated-shaped U-net (GSU-Net), was created to automatically segment the sellar region into 8 classes. Five magnetic resonance imaging (MRI) features were extracted from the segmentation results, including tumor diameters, volume, optic chiasma height, Knosp grading system, and degree of internal carotid artery contact. The clinical usefulness of the proposed methods was evaluated by the diagnostic accuracy of the tumor consistency. Results: A total of 163 patients with confirmed pituitary adenoma were included as the first group and were randomly divided into a training data set and test data set (131 and 32 patients, respectively). Fifty patients with confirmed acromegaly were included as the second group. The Dice coefficient of pituitary adenoma in important image slices was 0.940. The proposed methods achieved accuracies of more than 80% for the prediction of 5 invasive-related MRI features. Methods derived from the automatic segmentation showed better performance than original methods and achieved areas under the curve of 0.840 and 0.920 for clinical models and radiomics models, respectively. Conclusion: The proposed methods could automatically segment the sellar region and extract features with high accuracy. The outstanding performance of the prediction of the tumor consistency indicates the methods’ clinical usefulness for supporting neurosurgeons in judging patients’ conditions, predicting prognosis, and other downstream tasks during the preoperative period.

Background: Prediction of radiotherapeutic response before radiotherapy could help determine individual treatment strategies for patients with acromegaly. Objective: To develop and validate a machine-learning-based multiparametric MRI radiomics model to non-invasively predict radiotherapeutic response in patients with acromegaly. Methods: This retrospective study included 57 acromegaly patients who underwent postoperative radiotherapy between January 2008 and January 2016. Manual lesion segmentation and radiomics analysis were performed on each pituitary adenoma, and 1561 radiomics features were extracted from each sequence. A radiomics signature was built with a support vector machine using leave-one-out cross-validation for feature selection. Multivariable logistic regression analysis was used to select appropriate clinicopathological features to construct a clinical model, which was then combined with the radiomics signature to construct a radiomics model. The performance of this radiomic model was assessed using receiver operating characteristics (ROC) analysis and its calibration, discriminating ability, clinical usefulness. Results: At 3-years after radiotherapy, 25 patients had achieved remission and 32 patients had not. The clinical model incorporating seven clinical features had an area under the ROC (AUC) of 0.86 for predicting radiotherapeutic response, and performed better than any single clinical feature. The radiomics signature constructed with six radiomics features had a significantly higher AUC of 0.92. The radiomics model showed good discrimination abilities and calibration, with an AUC of 0.96. Decision curve analysis confirmed the clinical utility of the radiomics model. Conclusion: Using pre-radiotherapy clinical and MRI data, we developed a radiomics model with favorable performance for individualized non-invasive prediction of radiotherapeutic response, which may help in identifying acromegaly patients who are likely to benefit from radiotherapy.

Background: Prediction of the tumor consistency before surgery is of vital importance to determine individualized therapeutic schemes for patients with acromegaly. The present study was performed to noninvasively predict the tumor consistency based on magnetic resonance imaging and radiomics analysis. Methods: In total, 158 patients with acromegaly were randomized into the primary cohort (n = 100) and validation cohort (n = 58). The consistency of the tumor was classified as soft or firm according to the neurosurgeon’s evaluation. The critical radiomics features were determined using the elastic net feature selection algorithm, and the radiomics signature was constructed. The most valuable clinical characteristics were then selected based on the multivariable logistic regression analysis. Next, a radiomics model was developed using the radiomics signature and clinical characteristics, and 30 patients with acromegaly were recruited for prospective validation of the radiomics model. The model’s performance was evaluated based on the receiver operating characteristic (ROC) curve, area under the ROC curve (AUC), accuracy, and other associated classification measures. Its calibration, discriminating capacity, and clinical usefulness were also evaluated. Results: The radiomics signature established according to four radiomics characteristics screened in the primary cohort exhibited excellent discriminatory capacity in the validation cohort. The radiomics model, which incorporated both the radiomics signature and Knosp grade, displayed favorable discriminatory capacity and calibration, and the AUC was 0.83 (95% confidence interval, 0.81–0.85) and 0.81 (95% confidence interval, 0.78–0.83) in the primary and validation cohorts, respectively. Furthermore, compared with the clinical characteristics, the as-constructed radiomics model more effectively predicted the tumor consistency in patients with acromegaly. Moreover, the prospective validation and decision curve analysis suggested that the radiomics model was clinically useful. Conclusions: This radiomics model can assist neurosurgeons in predicting the tumor consistency in patients with acromegaly before surgery and facilitate the determination of individualized therapeutic schemes.

Purpose: This pilot study investigated the performance of C-X-C motif chemokine receptor 4 (CXCR4) molecular imaging (⁶⁸Ga-pentixafor PET/CT) in Cushing syndrome (CS) and the correlation between CXCR4 signaling interactions and glucose metabolism in adrenocorticotropin-cortisol pathway. Methods: We retrospectively evaluated 31 patients (16 patients with CS and 15 patients with nonfunctioning pituitary or adrenal adenomas). All patients underwent ⁶⁸Ga-pentixafor PET/CT, and 11 with pituitary adenoma also underwent ¹⁸F-FDG PET/CT. The diagnosis accuracy of ⁶⁸Ga-pentixafor PET/CT was calculated. The correlation between radiouptake along the pituitary-adrenal axis and hormone levels was calculated. Results: Patients with Cushing disease characterized a focal uptake in adrenocorticotropic hormone–producing pituitary adenoma (ACTH-PA). In ACTH-independent CS, there was increased uptake of ⁶⁸Ga-pentixafor in adrenal lesions but not in the pituitary fossa. The nonfunctioning pituitary or adrenal adenomas showed negative ⁶⁸Ga-pentixafor signal. The one patient with metastatic ectopic ACTH syndrome had multiple ⁶⁸Ga-pentixafor–avid lesions. Using the threshold of SUV_max >8.5 in the adrenal lesions, the sensitivity and specificity of ⁶⁸Ga-pentixafor PET/CT to diagnose cortisol-producing adenoma were 100% and 84.9%. A cutoff SUV_max value of 3.0 on ⁶⁸Ga-pentixafor PET/CT had 100% sensitivity and specificity for differentiating ACTH-PA. The corresponding hormone level was significantly correlated with uptake of ⁶⁸Ga-pentixafor in pituitary adenoma and adrenal tissue but not with glucose metabolism. Conclusion: We have characterized the performance of ⁶⁸Ga-pentixafor in different subtypes of CS. ⁶⁸Ga-pentixafor PET/CT is promising in the differential diagnosis of both ACTH-independent and ACTH-dependent CS. Activated CXCR4 molecular signaling along the pituitary-adrenal axis was found in patients with Cushing disease.

Background: Deep learning has the potential to assist the medical diagnostic process. We aimed to identify facial anomalies associated with endocrinal disorders using a deep-learning approach to facilitate the process of diagnosis and follow-up. Methods: We collected facial images of patients with hypercortisolism and acromegaly, and we augmented these images with additional negative samples from public databases. A model with a pretrained deep-learning network was constructed to automatically identify these hypersecretion statuses based on characteristic facial changes. We compared its performance to that of endocrine experts and further investigated key factors upon which the best performing model focused. Findings: The model achieved areas under the receiver operating characteristic curve of 0.9647 (Cushing’s syndrome) and 0.9556 (acromegaly), accuracies of 0.9593 (Cushing’s syndrome) and 0.9479 (acromegaly), and recalls of 0.7593 (Cushing’s syndrome) and 0.8089 (acromegaly). It performed better than any level of our endocrine experts. Furthermore, the regions of interest on the part of the machine were primarily the same as those upon which the humans focused. Interpretation: Our findings suggest that the deep-learning model learned the facial characters based merely on labeled data without learning prerequisite medical knowledge, and its performance was comparable with professional medical practitioners. The model has the potential to assist in the diagnosis and follow-up of these hypersecretion statuses.