Adithya Narayan

We introduce a systematic toolbox for diagnosing monocular depth estimation models by discovering failure cases arising from camera viewpoint changes. Our method employs a differentiable rendering pipeline to adversarially optimize camera extrinsics, sampling semantically meaningful but challenging viewpoints. We construct a dataset of complex 3D indoor scenes and evaluate four state-of-the-art depth estimation models, showing that our framework exposes consistent weaknesses and establishes a new benchmark for robustness analysis.

In this study, we propose a DL based segmentation framework for the automated segmentation of 10 key fetal brain structures from 2 axial planes from fetal brain USG images (2D). We developed a custom U-Net variant that uses inceptionv4 block as a feature extractor and leverages custom domain-specific data augmentation. Quantitatively, the mean (10 structures; test sets 1/2/3/4) Dice-coefficients were: 0.827, 0.802, 0.731, 0.783. Irrespective of the USG device/center, the DL segmentations were qualitatively comparable to their manual segmentations. The proposed DL system offered a promising and generalizable performance (multi-centers, multi-device) and also presents evidence in support of device-induced variation in image quality (a challenge to generalizibility) by using UMAP analysis.

We leveraged clinically relevant biometric constraints (relationship between caliper points) and domain-relevant data augmentation to improve the accuracy of a U-Net DL model (trained/tested on: 596 images, 473 subjects/143 images, 143 subjects). We performed multiple experiments demonstrating the effect of the DL backbone, data augmentation, generalizability and benchmarked against a recent state-of-the-art approach through extensive clinical validation (DL vs. 7 experienced clinicians). For all cases, the mean errors in the placement of the individual caliper points and the computed biometry were comparable to error rates among clinicians. The clinical translation of the proposed framework can assist novice users from low-resource settings in the reliable and standardized assessment of fetal brain sonograms.

We retrospectively obtained 4,190 two-dimensional (2D) ultrasonography (USG) images (1349 pregnancies; TV + TC images) from 3 centres (2 tertiary referral centre [TRC 1,2] + 1 routine imaging centre [RIC]) using 6 ultrasound (USG) devices (GE Voluson: P8,P6,E8,E10,S10; Samsung: HERA W10). A custom U-Net was trained (2744 images from TRC 1 [E8, S10]) on 2D fetal brain images (TV + TC images) and their corresponding manual segmentations to segment 10 key fetal structures (TV + TC planes). We assessed the robustness (operator & centre variability) and generalisability (across devices) of the proposed approach across 4 independent (unseen) test sets. Test set 1 (TRC 1, trained devices): 718 images (E8, S10); test 2 (TRC 1, unseen devices): 192 images (HERA W10, P6, E10); test set 3 (TRC 2, trained device): 378 images (E8), and test set 4 (RIC, unseen device): 158 images (P8). The segmentation performance was qualitatively and quantitatively (Dice coefficient [DC]) assessed.