Can AI distinguish a bone radiograph from photos of flowers or cars? Evaluation of bone age deep learning model on inappropriate data inputs

Abstract

Objective To evaluate the behavior of a publicly available deep convolutional neural network (DCNN) bone age algorithm when presented with inappropriate data inputs in both radiological and non-radiological domains. Methods We evaluated a publicly available DCNN-based bone age application. The DCNN was trained on 12,612 pediatric hand radiographs and won the 2017 RSNA Pediatric Bone Age Challenge (concordance of 0.991 with radiologist groundtruth). We used the application to analyze 50 left-hand radiographs (appropriate data inputs) and seven classes of inappropriate data inputs in radiological (i.e., chest radiographs) and non-radiological (i.e., image of street numbers) domains. For each image, we noted if (1) the application distinguished between appropriate and inappropriate data inputs and (2) inference time per image. Mean inference times were compared using ANOVA. Results The 16Bit Bone Age application calculated bone age for all pediatric hand radiographs with mean inference time of 1.1 s. The application did not distinguish between pediatric hand radiographs and inappropriate image types, including radiological and non-radiological domains. The application inappropriately calculated bone age for all inappropriate image types, with mean inference time of 1.1 s for all categories (p = 1). Conclusion A publicly available DCNN-based bone age application failed to distinguish between appropriate and inappropriate data inputs and calculated bone age for inappropriate images. The awareness of inappropriate outputs based on inappropriate DCNN input is important if tasks such as bone age determination are automated, emphasizing the need for appropriate oversight at the data input and verification stage to avoid unrecognized erroneous results.

Accuracy and self-validation of automated bone age determination

“…AI systems are new tools in radiology, and understandably doctors are unsure about their limitations. As an example, a recent paper 21 examined a deep learning bone age assessment method on a wide range of images, including shoulder and thorax images, and the method processed these images with no warning and gave completely wrong bone ages – the authors were critical about the system’s total lack of understanding of its own limitation. This study demonstrates that one cannot expect that radiologists can judge whether images are within the range of the AI method’s validity – and while it seems easy to guide users to not send non-hand images to the system, there are less obvious limitations: What if the image grey scale is inverted, if there are two hands in the image, if the pose of the hand is non-standard, if some fingers are missing, if there is a very large margin around the hand, etc.?…”

Can artificial intelligence pass the Fellowship of the Royal College of Radiologists examination? Multi-reader diagnostic accuracy study

“… 15 Nevertheless, the artificial intelligence should analyse only radiographs that it has been trained to interpret to avoid erroneous anomalies being highlighted on those deemed non-interpretable, which occurred in one case in this study (and has been known to occur with other artificial intelligence tools). 16 …”

Radiomics and Deep Learning for Disease Detection in Musculoskeletal Radiology

“…Most DL and DCNN architectures in musculoskeletal radiology have been applied to radiographs for fracture detection, osteoarthritis grading, bone age assessment, quantification tasks, and characterization of orthopedic implants. [42][43][44][45][46][47] Although fewer studies have been performed on MRI and CTof the musculoskeletal system, 48 an increasing number of studies describe DL and DCNN methods for MRI-or CT-based image reconstruction, tissue segmentation, and musculoskeletal disease detection. 42,49…”