Postoperative vault prediction for phakic implantable collamer lens surgery: LASSO formulas.

PURPOSE: To develop and evaluate reliable formulas for predicting postoperative vault more accurately after implantable collamer lens (ICL) surgery in a White patient population with varying degrees of ametropia. SETTING: Private clinical practice. DESIGN: Retrospective analysis on dataset split into a separate training and test set. METHODS: 115 eyes of 59 patients were used to train regression models predicting postoperative vault based on anterior segment optical coherence tomography (OCT) parameters (Least Absolute Shrinkage and Selection Operator [LASSO]-OCT formula), ocular biometry data (LASSO-Biometry formula), or data from both devices (LASSO-Full formula). The performance of these models was evaluated against the manufacturer's nomogram (Online Calculation and Ordering System [OCOS]) and Nakamura 1 (NK1) and 2 (NK2) formulas on a matched separate test set of 37 eyes of 19 patients. RESULTS: The mean preoperative spherical equivalent was -5.32 ± 3.37 (range: +3.75 to -17.375 diopters). The mean absolute errors of the estimated vs achieved postoperative vault for the LASSO-Biometry, LASSO-OCT, and LASSO-Full formulas were 144.1 ± 107.9 µm, 145.6 ± 100.6 µm, and 132.0 ± 86.6 µm, respectively. These results were significantly lower compared with the OCOS, NK1, and NK2 formulas ( P < .006). Postoperative vault could be estimated within 500 µm in 97.3% (LASSO-Biometry) to 100% of cases (LASSO-OCT and LASSO-Full). CONCLUSIONS: The LASSO suite provided a set of powerful, reproducible yet convenient ICL sizing formulas with state-of-the-art performance in White patients, including those with low to moderate degrees of myopia. The calculator can be accessed at http://icl.emmetropia.be .

Machine learning for filtering out false positive grey matter atrophies in single subject voxel based morphometry: A simulation based study.

Single subject VBM (SS-VBM), has been used as an alternative tool to standard VBM for single case studies. However, it has the disadvantage of producing an excessively large number of false positive detections. In this study we propose a machine learning technique widely used for automated data classification, namely Support Vector Machine (SVM), to refine the findings produced by SS-VBM. A controlled set of experiments was conducted to evaluate the proposed approach using three-dimensional T1 MRI scans from control subjects collected from the publicly available IXI dataset. The scans were artificially atrophied at different locations and with different sizes to mimic the behavior of neurological disorders. Results empirically demonstrated that the proposed method is able to significantly reduce the amount of false positive clusters (p < 0.05), with no statistical differences in the true positive findings (p > 0.05). This evidence was observed to be consistent for different atrophied areas and sizes of atrophies. This approach could be potentially be applied to alleviate the intensive manual analysis that radiologists and clinicians have to perform to filter out miss-detections of SS-VBM, increasing its usability for image reading.

Automated lumen segmentation using multi-frame convolutional neural networks in intravascular ultrasound datasets.

Aims: Assessment of minimum lumen areas in intravascular ultrasound (IVUS) pullbacks is time-consuming and demands adequately trained personnel. In this work, we introduce a novel and fully automated pipeline to segment the lumen boundary in IVUS datasets. Methods and results: First, an automated gating is applied to select end-diastolic frames and bypass saw-tooth artefacts. Second, within a machine learning (ML) environment, we automatically segment the lumen boundary using a multi-frame (MF) convolutional neural network (MFCNN). Finally, we use the theory of Gaussian processes (GPs) to regress the final lumen boundary. The dataset consisted of 85 IVUS pullbacks (52 patients). The dataset was partitioned at the pullback-level using 73 pullbacks for training (20 586 frames), 6 pullbacks for validation (1692 frames), and 6 for testing (1692 frames). The degree of overlapping, between the ground truth and ML contours, median (interquartile range, IQR) systematically increased from 0.896 (0.874-0.933) for MF1 to 0.925 (0.911-0.948) for MF11. The median (IQR) of the distance error was also reduced from 3.83 (2.94-4.98)% for MF1 to 3.02 (2.25-3.95)% for MF11-GP. The corresponding median (IQR) in the lumen area error remained between 5.49 (2.50-10.50)% for MF1 and 5.12 (2.15-9.00)% for MF11-GP. The dispersion in the relative distance and area errors consistently decreased as we increased the number of frames, and also when the GP regressor was coupled to the MFCNN output. Conclusion: These results demonstrate that the proposed ML approach is suitable to effectively segment the lumen boundary in IVUS scans, reducing the burden of costly and time-consuming manual delineation.