Beyond Size and Class Balance: Alpha as a New Dataset Quality Metric for Deep Learning

In deep learning, achieving high performance on image classification tasks requires diverse training sets. However, dataset diversity is incompletely understood. The current best practice is to try to maximize dataset size and class balance. Yet large, class-balanced datasets are not guaranteed to be diverse: images can still be arbitrarily similar. We hypothesized that, for a given model architecture, better model performance can be achieved by maximizing dataset diversity more directly. This could open a path for performance improvement without additional computational resources or architectural advances. To test this hypothesis, we introduce a comprehensive framework of diversity measures, developed in ecology, that generalizes familiar quantities like Shannon entropy by accounting for similarities and differences among images. (Dataset size and class balance emerge from this framework as special cases.) By analyzing thousands of subsets from seven medical datasets representing ultrasound, X-ray, CT, and pathology images, we found that the best correlates of performance were not size or class balance but $A$ -- ``big alpha'' -- a set of generalized entropy measures interpreted as the effective number of image-class pairs in the dataset, after accounting for similarities among images. One of these, $A_0$, explained 67\% of the variance in balanced accuracy across all subsets, vs. 54\% for class balance and just 39\% for size. The best pair was size and $A_1$ (79\%), which outperformed size and class balance (74\%). $A$ performed best for subsets from individual datasets as well as across datasets, supporting the generality of these results. We propose maximizing $A$ as a potential new way to improve the performance of deep learning in medical imaging.

Are foundation models efficient for medical image segmentation?

Foundation models are experiencing a surge in popularity. The Segment Anything model (SAM) asserts an ability to segment a wide spectrum of objects but required supervised training at unprecedented scale. We compared SAM's performance (against clinical ground truth) and resources (labeling time, compute) to a modality-specific, label-free self-supervised learning (SSL) method on 25 measurements for 100 cardiac ultrasounds. SAM performed poorly and required significantly more labeling and computing resources, demonstrating worse efficiency than SSL.

Label-free segmentation from cardiac ultrasound using self-supervised learning

Background: Segmentation and measurement of cardiac chambers is critical in echocardiography but is also laborious and poorly reproducible. Neural networks can assist, but supervised approaches require the same laborious manual annotations, while unsupervised approaches have fared poorly in ultrasound to date. Objectives: We built a pipeline for self-supervised (no manual labels required) segmentation of cardiac chambers, combining computer vision, clinical domain knowledge, and deep learning. Methods: We trained on 450 echocardiograms (145,000 images) and tested on 8,393 echocardiograms (4,476,266 images; mean age 61 years, 51% female), using the resulting segmentations to calculate structural and functional measurements. We also tested our pipeline against external images from an additional 10,030 patients (20,060 images) with available manual tracings of the left ventricle. Results: r2 between clinically measured and pipeline-predicted measurements were similar to reported inter-clinician variation for LVESV and LVEDV (pipeline vs. clinical r2= 0.74 and r2=0.65, respectively), LVEF and LV mass (r2= 0.46 and r2=0.54), left and right atrium volumes (r2=0.7 and r2=0.6), and right ventricle area (r2=0.47). When binarized into normal vs. abnormal categories, average accuracy was 0.81 (range 0.71-0.95). A subset of the test echocardiograms (n=553) had corresponding cardiac MRI; correlation between pipeline and CMR measurements was similar to that between clinical echocardiogram and CMR. Finally, in the external dataset, our pipeline accurately segments the left ventricle with an average Dice score of 0.83 (95% CI 0.83). Conclusions: Our results demonstrate a human-label-free, valid, and scalable method for segmentation from ultrasound, a noisy but globally important imaging modality.

Domain-guided data augmentation for deep learning on medical imaging

While domain-specific data augmentation can be useful in training neural networks for medical imaging tasks, such techniques have not been widely used to date. Here, we test whether domain-specific data augmentation is useful for medical imaging using a well-benchmarked task: view classification on fetal ultrasound FETAL-125 and OB-125 datasets. We found that using a context-preserving cut-paste strategy, we could create valid training data as measured by performance of the resulting trained model on the benchmark test dataset. When used in an online fashion, models trained on this data performed similarly to those trained using traditional data augmentation (FETAL-125 F-score 85.33+/-0.24 vs 86.89+/-0.60, p-value 0.0139; OB-125 F-score 74.60+/-0.11 vs 72.43+/-0.62, p-value 0.0039). Furthermore, the ability to perform augmentations during training time, as well as the ability to apply chosen augmentations equally across data classes, are important considerations in designing a bespoke data augmentation. Finally, we provide open-source code to facilitate running bespoke data augmentations in an online fashion. Taken together, this work expands the ability to design and apply domain-guided data augmentations for medical imaging tasks.

Deep-learning models improve on community-level diagnosis for common congenital heart disease lesions

Prenatal diagnosis of tetralogy of Fallot (TOF) and hypoplastic left heart syndrome (HLHS), two serious congenital heart defects, improves outcomes and can in some cases facilitate in utero interventions. In practice, however, the fetal diagnosis rate for these lesions is only 30-50 percent in community settings. Improving fetal diagnosis of congenital heart disease is therefore critical. Deep learning is a cutting-edge machine learning technique for finding patterns in images but has not yet been applied to prenatal diagnosis of congenital heart disease. Using 685 retrospectively collected echocardiograms from fetuses 18-24 weeks of gestational age from 2000-2018, we trained convolutional and fully-convolutional deep learning models in a supervised manner to (i) identify the five canonical screening views of the fetal heart and (ii) segment cardiac structures to calculate fetal cardiac biometrics. We then trained models to distinguish by view between normal hearts, TOF, and HLHS. In a holdout test set of images, F-score for identification of the five most important fetal cardiac views was 0.95. Binary classification of unannotated cardiac views of normal heart vs. TOF reached an overall sensitivity of 75% and a specificity of 76%, while normal vs. HLHS reached a sensitivity of 100% and specificity of 90%, both well above average diagnostic rates for these lesions. Furthermore, segmentation-based measurements for cardiothoracic ratio (CTR), cardiac axis (CA), and ventricular fractional area change (FAC) were compatible with clinically measured metrics for normal, TOF, and HLHS hearts. Thus, using guideline-recommended imaging, deep learning models can significantly improve detection of fetal congenital heart disease compared to the common standard of care.

Fast and accurate classification of echocardiograms using deep learning

Echocardiography is essential to modern cardiology. However, human interpretation limits high throughput analysis, limiting echocardiography from reaching its full clinical and research potential for precision medicine. Deep learning is a cutting-edge machine-learning technique that has been useful in analyzing medical images but has not yet been widely applied to echocardiography, partly due to the complexity of echocardiograms' multi view, multi modality format. The essential first step toward comprehensive computer assisted echocardiographic interpretation is determining whether computers can learn to recognize standard views. To this end, we anonymized 834,267 transthoracic echocardiogram (TTE) images from 267 patients (20 to 96 years, 51 percent female, 26 percent obese) seen between 2000 and 2017 and labeled them according to standard views. Images covered a range of real world clinical variation. We built a multilayer convolutional neural network and used supervised learning to simultaneously classify 15 standard views. Eighty percent of data used was randomly chosen for training and 20 percent reserved for validation and testing on never seen echocardiograms. Using multiple images from each clip, the model classified among 12 video views with 97.8 percent overall test accuracy without overfitting. Even on single low resolution images, test accuracy among 15 views was 91.7 percent versus 70.2 to 83.5 percent for board-certified echocardiographers. Confusional matrices, occlusion experiments, and saliency mapping showed that the model finds recognizable similarities among related views and classifies using clinically relevant image features. In conclusion, deep neural networks can classify essential echocardiographic views simultaneously and with high accuracy. Our results provide a foundation for more complex deep learning assisted echocardiographic interpretation.