Improved breast cancer risk models enable targeted screening strategies that achieve earlier detection and less screening harm than existing guidelines. To bring deep learning risk models to clinical practice, we need to further refine their accuracy, validate them across diverse populations, and demonstrate their potential to improve clinical workflows. We developed Mirai, a mammography-based deep learning model designed to predict risk at multiple timepoints, leverage potentially missing risk factor information, and produce predictions that are consistent across mammography machines. Mirai was trained on a large dataset from Massachusetts General Hospital (MGH) in the United States and tested on held-out test sets from MGH, Karolinska University Hospital in Sweden, and Chang Gung Memorial Hospital (CGMH) in Taiwan, obtaining C-indices of 0.76 (95% confidence interval, 0.74 to 0.80), 0.81 (0.79 to 0.82), and 0.79 (0.79 to 0.83), respectively. Mirai obtained significantly higher 5-year ROC AUCs than the Tyrer-Cuzick model (P < 0.001) and prior deep learning models Hybrid DL (P < 0.001) and Image-Only DL (P < 0.001), trained on the same dataset. Mirai more accurately identified high-risk patients than prior methods across all datasets. On the MGH test set, 41.5% (34.4 to 48.5) of patients who would develop cancer within 5 years were identified as high risk, compared with 36.1% (29.1 to 42.9) by Hybrid DL (P = 0.02) and 22.9% (15.9 to 29.6) by the Tyrer-Cuzick model (P < 0.001).

PURPOSE: Low-dose computed tomography (LDCT) for lung cancer screening is effective, although most eligible people are not being screened. Tools that provide personalized future cancer risk assessment could focus approaches toward those most likely to benefit. We hypothesized that a deep learning model assessing the entire volumetric LDCT data could be built to predict individual risk without requiring additional demographic or clinical data. METHODS: We developed a model called Sybil using LDCTs from the National Lung Screening Trial (NLST). Sybil requires only one LDCT and does not require clinical data or radiologist annotations; it can run in real time in the background on a radiology reading station. Sybil was validated on three independent data sets: a heldout set of 6,282 LDCTs from NLST participants, 8,821 LDCTs from Massachusetts General Hospital (MGH), and 12,280 LDCTs from Chang Gung Memorial Hospital (CGMH, which included people with a range of smoking history including nonsmokers). RESULTS: Sybil achieved area under the receiver-operator curves for lung cancer prediction at 1 year of 0.92 (95% CI, 0.88 to 0.95) on NLST, 0.86 (95% CI, 0.82 to 0.90) on MGH, and 0.94 (95% CI, 0.91 to 1.00) on CGMH external validation sets. Concordance indices over 6 years were 0.75 (95% CI, 0.72 to 0.78), 0.81 (95% CI, 0.77 to 0.85), and 0.80 (95% CI, 0.75 to 0.86) for NLST, MGH, and CGMH, respectively. CONCLUSION: Sybil can accurately predict an individual's future lung cancer risk from a single LDCT scan to further enable personalized screening. Future study is required to understand Sybil's clinical applications. Our model and annotations are publicly available. [Media: see text]

The analysis of the backscattered statistics using the Nakagami parameter is an emerging ultrasound technique for assessing hepatic steatosis and fibrosis. Previous studies indicated that the echo amplitude distribution of a normal liver follows the Rayleigh distribution (the Nakagami parameter m is close to 1). However, using different frequencies may change the backscattered statistics of normal livers. This study explored the frequency dependence of the backscattered statistics in human livers and then discussed the sources of ultrasound scattering in the liver. A total of 30 healthy participants were enrolled to undergo a standard care ultrasound examination on the liver, which is a natural model containing diffuse and coherent scatterers. The liver of each volunteer was scanned from the right intercostal view to obtain image raw data at different central frequencies ranging from 2 to 3.5 MHz. Phantoms with diffuse scatterers only were also made to perform ultrasound scanning using the same protocol for comparisons with clinical data. The Nakagami parameter–frequency correlation was evaluated using Pearson correlation analysis. The median and interquartile range of the Nakagami parameter obtained from livers was 1.00 (0.98–1.05) for 2 MHz, 0.93 (0.89–0.98) for 2.3 MHz, 0.87 (0.84–0.92) for 2.5 MHz, 0.82 (0.77–0.88) for 3.3 MHz, and 0.81 (0.76–0.88) for 3.5 MHz. The Nakagami parameter decreased with the increasing central frequency (r = −0.67, p < 0.0001). However, the effect of ultrasound frequency on the statistical distribution of the backscattered envelopes was not found in the phantom results (r = −0.147, p = 0.0727). The current results demonstrated that the backscattered statistics of normal livers is frequency-dependent. Moreover, the coherent scatterers may be the primary factor to dominate the frequency dependence of the backscattered statistics in a liver.

The purpose of this work was to evaluate the performance of an existing commercially available artificial intelligence (AI) software system in differentiating malignant and benign lung nodules. The AI tool consisted of a vessel-suppression function and a deep-learning-based computer-aided-detection (VS-CAD) analyzer. Fifty patients (32 females, mean age 52 years) with 75 lung nodules (47 malignant and 28 benign) underwent low-dose computed tomography (LDCT) followed by surgical excision and the pathological analysis of their 75 nodules within a 3 month time frame. All 50 cases were then processed by the AI software to generate corresponding VS images and CAD outcomes. All 75 pathologically proven lung nodules were well delineated by vessel-suppressed images. Three (6.4%) of the 47 lung cancer cases, and 11 (39.3%) of the 28 benign nodules were ignored and not detected by the AI without showing a CAD analysis summary. The AI system/radiologists produced a sensitivity and specificity (shown in %) of 93.6/89.4 and 39.3/82.1 in distinguishing malignant from benign nodules, respectively. AI sensitivity was higher than that of radiologists, though not statistically significant (p = 0.712). Specificity obtained by the radiologists was significantly higher than that of the VS-CAD AI (p = 0.003). There was no significant difference between the malignant and benign lesions with respect to age, gender, pure ground-glass pattern, the diameter and location of the nodules, or nodules <6 vs. ≥6 mm. However, more part-solid nodules were proven to be malignant than benign (90.9% vs. 9.1%), and more solid nodules were proven to be benign than malignant (86.7% vs. 13.3%) with statistical significance (p = 0.001 and <0.001, respectively). A larger cohort and prospective study are required to validate the AI performance.