Justifying diagnosis decisions by deep neural networks

Abstract

An integrated approach is proposed across visual and textual data to both determine and justify a medical diagnosis by a neural network. As deep learning techniques improve, interest grows to apply them in medical applications. To enable a transition to workflows in a medical context that are aided by machine learning, the need exists for such algorithms to help justify the obtained outcome so human clinicians can judge their validity. In this work, deep learning methods are used to map a frontal X-Ray image to a continuous textual representation. This textual representation is decoded into a diagnosis and the associated textual justification that will help a clinician evaluate the outcome. Additionally, more explanatory data is provided for the diagnosis by generating a realistic X-Ray that belongs to the nearest alternative diagnosis. With a clinical expert opinion study on a subset of the X-Ray data set from the Indiana University hospital network, we demonstrate that our justification mechanism significantly outperforms existing methods that use saliency maps. While performing multi-task training with multiple loss functions, our method achieves excellent diagnosis accuracy and captioning quality when compared to current state-of-the-art single-task methods.

Recent advances of HCI in decision-making tasks for optimized clinical workflows and precision medicine

“…The proposed approach was beneficial for gaining an in-depth understanding of the sense-making process during this critical task, as well as for identifying design requirements for better sense-making support. In [59], Deep Learning (DL) techniques were exploited to generate a diagnosis as textual representation from a frontal X-Ray image. Moreover, realistic X-Ray images related to the nearest alternative diagnosis were generated.…”

Fusion High-Resolution Network for Diagnosing ChestX-ray Images

“…in which O represents the fusion feature, W represents the mapping matrix that is used to map the high-dimensional fusion feature to a low-dimensional probability distribution representing the disease information and p i (1 ≤ i ≤ 14) represents the probability of identifying the ith disease [48].…”

A Survey on Deep Learning and Explainability for Automatic Report Generation from Medical Images

“…Used by papers DenseNet [59] [ 15,37,83,84,87,90,143,147,154] ResNet [51] [ 39,43,47,60,65,87,92,136,145,146,148] VGG [116] [7, 35, 39, 50, 66, 74, 85, 93, 95, 149, 150] Faster R-CNN [106] [74, 149] Inception V3 [124] [117] GoogLeNet [123] [114] MobileNet V2 [58] [48] SRN [158] [43] U-Net [110] [122] EcNet (*) [155] FCN + shallow CNN (*) [125] RGAN (*) [46] StackGAN [151] (slightly modified version) (*) [120] CNN (*) [120, 126] CNN (unspecified architecture) [140,142] Table 4. Summary of convolutional neural network architectures used in the literature.…”

“…RGAN, proposed by Han et al [46], is a novel architecture that follows the generative adversarial network (GAN) [40] approach, with a generative module comprising the encoder and decoder parts of an atrous convolution autoencoder (ACAE) with a spatial LSTM between them. Similarly, Spinks and Moens [120] used a slightly modified version of a StackGAN [151] to learn the mapping from report encoding to chest X-ray images, and a custom CNN to learn the inverse mapping. Both are trained together, but only the latter is part of the report generation pipeline during inference.…”