<b><i>Background and Aims:</i></b> Current prediction models for early recurrence of hepatocellular carcinoma (HCC) after surgical resection remain unsatisfactory. The aim of this study was to develop evolutionary learning-derived prediction models with interpretability using both clinical and radiomic features to predict early recurrence of HCC after surgical resection. <b><i>Methods:</i></b> Consecutive 517 HCC patients receiving surgical resection with available contrast-enhanced computed tomography (CECT) images before resection were retrospectively enrolled. Patients were randomly assigned to a training set (<i>n</i> = 362) and a test set (<i>n</i> = 155) in a ratio of 7:3. Tumor segmentation of all CECT images including noncontrast phase, arterial phase, and portal venous phase was manually performed for radiomic feature extraction. A novel evolutionary learning-derived method called genetic algorithm for predicting recurrence after surgery of liver cancer (GARSL) was proposed to design prediction models for early recurrence of HCC within 2 years after surgery. <b><i>Results:</i></b> A total of 143 features, including 26 preoperative clinical features, 5 postoperative pathological features, and 112 radiomic features were used to develop GARSL preoperative and postoperative models. The area under the receiver operating characteristic curves (AUCs) for early recurrence of HCC within 2 years were 0.781 and 0.767, respectively, in the training set, and 0.739 and 0.741, respectively, in the test set. The accuracy of GARSL models derived from the evolutionary learning method was significantly better than models derived from other well-known machine learning methods or the early recurrence after surgery for liver tumor (ERASL) preoperative (AUC = 0.687, <i>p</i> &#x3c; 0.001 vs. GARSL preoperative) and ERASL postoperative (AUC = 0.688, <i>p</i> &#x3c; 0.001 vs. GARSL postoperative) models using clinical features only. <b><i>Conclusion:</i></b> The GARSL models using both clinical and radiomic features significantly improved the accuracy to predict early recurrence of HCC after surgical resection, which was significantly better than other well-known machine learning-derived models and currently available clinical models.

Pompe disease is a hereditary neuromuscular disorder attributed to acid α-glucosidase deficiency, and accurately identifying this disease is essential. Our aim was to discriminate normal muscles from neuropathic muscles in children affected by Pompe disease using a texture-feature parametric imaging method that simultaneously considers microstructure and macrostructure. The study included 22 children aged 0.02–54 months with Pompe disease and six healthy children aged 2–12 months with normal muscles. For each subject, transverse ultrasound images of the bilateral rectus femoris and sartorius muscles were obtained. Gray-level co-occurrence matrix-based Haralick’s features were used for constructing parametric images and identifying neuropathic muscles: autocorrelation (AUT), contrast, energy (ENE), entropy (ENT), maximum probability (MAXP), variance (VAR), and cluster prominence (CPR). Stepwise regression was used in feature selection. The Fisher linear discriminant analysis was used for combination of the selected features to distinguish between normal and pathological muscles. The VAR and CPR were the optimal feature set for classifying normal and pathological rectus femoris muscles, whereas the ENE, VAR, and CPR were the optimal feature set for distinguishing between normal and pathological sartorius muscles. The two feature sets were combined to discriminate between children with and without neuropathic muscles affected by Pompe disease, achieving an accuracy of 94.6%, a specificity of 100%, a sensitivity of 93.2%, and an area under the receiver operating characteristic curve of 0.98 ± 0.02. The CPR for the rectus femoris muscles and the AUT, ENT, MAXP, and VAR for the sartorius muscles exhibited statistically significant differences in distinguishing between the infantile-onset Pompe disease and late-onset Pompe disease groups (p < 0.05). Texture-feature parametric imaging can be used to quantify and map tissue structures in skeletal muscles and distinguish between pathological and normal muscles in children or newborns.

Our study aimed to evaluate the utility of muscle ultrasound in newborn screening of infantile-onset Pompe disease (IOPD) and to establish a system of severity grading. We retrospectively selected 35 patients with initial low acid alpha-glucosidase (GAA) activity and collected data including muscle ultrasound features, GAA gene mutation, activity/performance, and pathological and laboratory findings. The echogenicity of 6 muscles (the bilateral vastus intermedius, rectus femoris, and sartorius muscles) was compared to that of epimysium on ultrasound and rated either 1 (normal), 2 (mildly increased), or 3 (obviously increased). These grades were used to divide patients into 3 groups. IOPD was present in none of the grade-1 patients, 5 of 9 grade-2 patients, and 5 of 5 grade-3 patients (P < .001). Comparing grade-2 plus grade-3 patients to grade-1 patients, muscle ultrasound detected IOPD with a sensitivity and specificity of 100.0% (95% confidence interval [CI]: 69.2%–100%) and 84.0% (95% CI: 63.9%–95.5%), respectively. The mean number of affected muscles was larger in grade-3 patients than in grade-2 patients (4.2 vs. 2.0, P = .005). Mean alanine transaminase (ALT), aspartate transaminase (AST), creatine kinase (CK), and lactate dehydrogenase (LDH) levels were differed significantly different between grade-3 and grade-1 patients (P < .001). Because it permits direct visualization of injured muscles, muscle ultrasound can be used to screen for IOPD. Our echogenicity grades of muscle injury also correlate well with serum levels of muscle-injury biochemical markers.

No abstract