Deep learning fetal ultrasound video model match human observers in biometric measurements

Abstract

Objective: This work investigates the use of deep convolutional neural networks (CNN) to automatically perform measurements of fetal body parts, including head circumference, biparietal diameter, abdominal circumference and femur length, and to estimate gestational age and fetal weight using fetal ultrasound videos. Approach: We developed a novel multi-task CNN-based spatio-temporal fetal US feature extraction and standard plane detection algorithm (called FUVAI) and evaluated the method on 50 freehand fetal US video scans. We compared FUVAI fetal biometric measurements with measurements made by five experienced sonographers at two time points separated by at least two weeks. Intra- and inter-observer variabilities were estimated. Main Results: We found that automated fetal biometric measurements obtained by FUVAI were comparable to the measurements performed by experienced sonographers The observed differences in measurement values were within the range of inter- and intra-observer variability. Moreover, analysis has shown that these differences were not statistically significant when comparing any individual medical expert to our model. Significance: We argue that FUVAI has the potential to assist sonographers who perform fetal biometric measurements in clinical settings by providing them with suggestions regarding the best measuring frames, along with automated measurements. Moreover, FUVAI is able perform these tasks in just a few seconds, which is a huge difference compared to the average of six minutes taken by sonographers. This is significant, given the shortage of medical experts capable of interpreting fetal ultrasound images in numerous countries.

Fetal biometry and amniotic fluid volume assessment end-to-end automation using Deep Learning

“…A similar study to ours compared the performances of a multi-task deep neural network (DNN) on FB assessment, testing it on 50 free-hand ultrasound videos with results comparable to those of trained sonographers. Our models outperformed the one described in the study (FUVAI) 14 when comparing proximity of the results showcased by the model vs sonographers expressed in MAE (table 2) even if the DICE score coe cients and IoU were lower for the same tasks potentially indicating a greater generalizability of our models. FUVAI's choice of standard biometry planes didn't rely on the quality of the plane but rather on the con dence of the model when selecting it; in other words, on how closely it resembled images from the training set which are not necessarily the best standard planes according to the ISUOG guidelines.…”

“…The FUVAI model is the closest one to our approach for end-to-end automated biometric assessment from cine-loops and showed similar performances to those of trained sonographers 14 .…”

“…We computed the MAE of each parameter using the open source FUVAI model developed by Plotka et al 14 and compared them with our approach (table 2).…”

“…Advances in deep learning (DL) applied to medical imaging have sparked interest in its application to measurement automation in fetal ultrasound, with studies showcasing human-level performances of DL models in standard plane classi cation and segmentation [14][15][16][17] . Most of them are retrospective "in silico" studies conducted on Caucasian populations on xed images except for a few exceptions 18 .…”

POTHER: Patch-Voted Deep Learning-Based Chest X-ray Bias Analysis for COVID-19 Detection

“…Multi-task neural network. Inspired by [15], [16], [17], we use an encoderdecoder based convolutional neural network (CNN) for simultaneous classification and segmentation of the lungs in the CXRs. Our network is U-Net-based with ImageNet pre-trained ResNet-50 as backbone encoder [6].…”