Radiomics of Preoperative Multi-Sequence Magnetic Resonance Imaging Can Improve the Predictive Performance of Microvascular Invasion in Hepatocellular Carcinoma

Background

The aim of the study is to demonstrate that radiomics of preoperative multi-sequence magnetic resonance imaging (MRI) can indeed improve the predictive performance of microvascular invasion (MVI) in hepatocellular carcinoma (HCC).

Methods

A total of 206 patients with pathologically confirmed HCC who underwent preoperative enhanced MRI were retrospectively recruited. Univariate and multivariate logistic regression analysis identified the independent clinicoradiologic predictors of MVI present and constituted the clinicoradiologic model. Recursive feature elimination (RFE) was applied to select radiomics features (extracted from six sequence images) and constructed the radiomics model. Clinicoradiologic model plus radiomics model formed the clinicoradiomics model. Five-fold cross-validation was used to validate the three models. Discrimination, calibration, and clinical utility were used to evaluate the performance. Net reclassification improvement (NRI) and integrated discrimination improvement (IDI) were used to compare the prediction accuracy between models.

Results

The clinicoradiologic model contained alpha-fetoprotein (AFP)_lg10, radiological capsule enhancement, enhancement pattern and arterial peritumoral enhancement, which were independent risk factors of MVI. There were 18 radiomics features related to MVI constructed the radiomics model. The mean area under the receiver operating curve (AUC) of clinicoradiologic, radiomics and clinicoradiomics model were 0.849, 0.925 and 0.950 in the training cohort and 0.846, 0.907 and 0.933 in the validation cohort, respectively. The three models' calibration curves fitted well, and decision curve analysis (DCA) confirmed the clinical usefulness. Compared with the clinicoradiologic model, the NRI of radiomics and clinicoradiomics model increased significantly by 0.575 and 0.825, respectively, and the IDI increased significantly by 0.280 and 0.398, respectively.

Conclusions

Radiomics of preoperative multi-sequence MRI can improve the predictive performance of MVI in HCC.

Intratumoral and peritumoral MRI radiomics nomogram for predicting parametrial invasion in patients with early-stage cervical adenocarcinoma and adenosquamous carcinoma

OBJECTIVE

To develop a comprehensive nomogram based on MRI intra- and peritumoral radiomics signatures and independent risk factors for predicting parametrial invasion (PMI) in patients with early-stage cervical adenocarcinoma (AC) and adenosquamous carcinoma (ASC).

METHODS

A total of 460 patients with IB to IIB cervical AC and ASC who underwent preoperative MRI examination and radical trachelectomy/hysterectomy were retrospectively enrolled and divided into primary, internal validation, and external validation cohorts. The original (Ori) and wavelet (Wav)-transform features were extracted from the volumetric region of interest of the tumour (ROI-T) and 3mm- and 5mm-peritumoral rings (ROI-3 and ROI-5), respectively. Then the Ori and Ori-Wav feature-based radiomics signatures from the tumour (RST) and 3 mm- and 5 mm-peritumoral regions (RS3 and RS5) were independently built and their diagnostic performances were compared to select the optimal ones. Finally, the nomogram was developed by integrating optimal intra- and peritumoral signatures and clinical independent risk factors based on multivariable logistic regression analysis.

RESULTS

FIGO stage, disruption of the cervical stromal ring on MRI (DCSRMR), parametrial invasion on MRI (PMIMR), and serum CA-125 were identified as independent risk factors. The nomogram constructed by integrating independent risk factors, Ori-Wav feature-based RST, and RS5 yielded AUCs of 0.874 (0.810-0.922), 0.885 (0.834-0.924), and 0.966 (0.887-0.995) for predicting PMI in the primary, internal and external validation cohorts, respectively. Furthermore, the nomogram was superior to radiomics signatures and clinical model for predicting PMI in three cohorts.

CONCLUSION

The nomogram can preoperatively, accurately, and noninvasively predict PMI in patients with early-stage cervical AC and ASC.

CLINICAL RELEVANCE STATEMENT

The nomogram can preoperatively, accurately, and noninvasively predict PMI and facilitate precise treatment decisions regarding chemoradiotherapy or radical hysterectomy in patients with early-stage cervical AC and ASC.

KEY POINTS

The accurate preoperative prediction of PMI in early-stage cervical AC and ASC can facilitate precise treatment decisions regarding chemoradiotherapy or radical hysterectomy. The nomogram integrating independent risk factors, Ori-Wav feature-based RST, and RS5 can preoperatively, accurately, and noninvasively predict PMI in early-stage cervical AC and ASC. The nomogram was superior to radiomics signatures and clinical model for predicting PMI in early-stage cervical AC and ASC.

A fully automated rib fracture detection system on chest CT images and its impact on radiologist performance

OBJECTIVE

To compare rib fracture detection and classification by radiologists using CT images with and without a deep learning model.

MATERIALS AND METHODS

A total of 8529 chest CT images were collected from multiple hospitals for training the deep learning model. The test dataset included 300 chest CT images acquired using a single CT scanner. The rib fractures were marked in the bone window on each CT slice by experienced radiologists, and the ground truth included 861 rib fractures. We proposed a heterogeneous neural network for rib fracture detection and classification consisting of a cascaded feature pyramid network and a classification network. The deep learning-based model was evaluated based on the external testing data. The precision rate, recall rate, F1-score, and diagnostic time of two junior radiologists with and without the deep learning model were computed, and the Chi-square, one-way analysis of variance, and least significant difference tests were used to analyze the results.

RESULTS

The use of the deep learning model increased detection recall and classification accuracy (0.922 and 0.863) compared with the radiologists alone (0.812 vs. 0.850). The radiologists achieved a higher precision rate, recall rate, and F1-score for fracture detection when using the deep learning model, at 0.943, 0.978, and 0.960, respectively. When using the deep learning model, the radiologist's reading time was decreased from 158.3 ± 35.7 s to 42.3 ± 6.8 s.

CONCLUSION

Radiologists achieved the highest performance in diagnosing and classifying rib fractures on CT images when assisted by the deep learning model.