Radiological age assessment based on clavicle ossification in CT: enhanced accuracy through deep learning.

BACKGROUND: Radiological age assessment using reference studies is inherently limited in accuracy due to a finite number of assignable skeletal maturation stages. To overcome this limitation, we present a deep learning approach for continuous age assessment based on clavicle ossification in computed tomography (CT). METHODS: Thoracic CT scans were retrospectively collected from the picture archiving and communication system. Individuals aged 15.0 to 30.0 years examined in routine clinical practice were included. All scans were automatically cropped around the medial clavicular epiphyseal cartilages. A deep learning model was trained to predict a person's chronological age based on these scans. Performance was evaluated using mean absolute error (MAE). Model performance was compared to an optimistic human reader performance estimate for an established reference study method. RESULTS: The deep learning model was trained on 4,400 scans of 1,935 patients (training set: mean age = 24.2 years ± 4.0, 1132 female) and evaluated on 300 scans of 300 patients with a balanced age and sex distribution (test set: mean age = 22.5 years ± 4.4, 150 female). Model MAE was 1.65 years, and the highest absolute error was 6.40 years for females and 7.32 years for males. However, performance could be attributed to norm-variants or pathologic disorders. Human reader estimate MAE was 1.84 years and the highest absolute error was 3.40 years for females and 3.78 years for males. CONCLUSIONS: We present a deep learning approach for continuous age predictions using CT volumes highlighting the medial clavicular epiphyseal cartilage with performance comparable to the human reader estimate.

ChatGPT makes medicine easy to swallow: an exploratory case study on simplified radiology reports.

OBJECTIVES: To assess the quality of simplified radiology reports generated with the large language model (LLM) ChatGPT and to discuss challenges and chances of ChatGPT-like LLMs for medical text simplification. METHODS: In this exploratory case study, a radiologist created three fictitious radiology reports which we simplified by prompting ChatGPT with "Explain this medical report to a child using simple language." In a questionnaire, we tasked 15 radiologists to rate the quality of the simplified radiology reports with respect to their factual correctness, completeness, and potential harm for patients. We used Likert scale analysis and inductive free-text categorization to assess the quality of the simplified reports. RESULTS: Most radiologists agreed that the simplified reports were factually correct, complete, and not potentially harmful to the patient. Nevertheless, instances of incorrect statements, missed relevant medical information, and potentially harmful passages were reported. CONCLUSION: While we see a need for further adaption to the medical field, the initial insights of this study indicate a tremendous potential in using LLMs like ChatGPT to improve patient-centered care in radiology and other medical domains. CLINICAL RELEVANCE STATEMENT: Patients have started to use ChatGPT to simplify and explain their medical reports, which is expected to affect patient-doctor interaction. This phenomenon raises several opportunities and challenges for clinical routine. KEY POINTS: • Patients have started to use ChatGPT to simplify their medical reports, but their quality was unknown. • In a questionnaire, most participating radiologists overall asserted good quality to radiology reports simplified with ChatGPT. However, they also highlighted a notable presence of errors, potentially leading patients to draw harmful conclusions. • Large language models such as ChatGPT have vast potential to enhance patient-centered care in radiology and other medical domains. To realize this potential while minimizing harm, they need supervision by medical experts and adaption to the medical field.

Automated localization of the medial clavicular epiphyseal cartilages using an object detection network: a step towards deep learning-based forensic age assessment.

BACKGROUND: Deep learning is a promising technique to improve radiological age assessment. However, expensive manual annotation by experts poses a bottleneck for creating large datasets to appropriately train deep neural networks. We propose an object detection approach to automatically annotate the medial clavicular epiphyseal cartilages in computed tomography (CT) scans. METHODS: The sternoclavicular joints were selected as structure-of-interest (SOI) in chest CT scans and served as an easy-to-identify proxy for the actual medial clavicular epiphyseal cartilages. CT slices containing the SOI were manually annotated with bounding boxes around the SOI. All slices in the training set were used to train the object detection network RetinaNet. Afterwards, the network was applied individually to all slices of the test scans for SOI detection. Bounding box and slice position of the detection with the highest classification score were used as the location estimate for the medial clavicular epiphyseal cartilages inside the CT scan. RESULTS: From 100 CT scans of 82 patients, 29,656 slices were used for training and 30,846 slices from 110 CT scans of 110 different patients for testing the object detection network. The location estimate from the deep learning approach for the SOI was in a correct slice in 97/110 (88%), misplaced by one slice in 5/110 (5%), and missing in 8/110 (7%) test scans. No estimate was misplaced by more than one slice. CONCLUSIONS: We demonstrated a robust automated approach for annotating the medial clavicular epiphyseal cartilages. This enables training and testing of deep neural networks for age assessment.

Deep learning in CT colonography: differentiating premalignant from benign colorectal polyps.

OBJECTIVES: To investigate the differentiation of premalignant from benign colorectal polyps detected by CT colonography using deep learning. METHODS: In this retrospective analysis of an average risk colorectal cancer screening sample, polyps of all size categories and morphologies were manually segmented on supine and prone CT colonography images and classified as premalignant (adenoma) or benign (hyperplastic polyp or regular mucosa) according to histopathology. Two deep learning models SEG and noSEG were trained on 3D CT colonography image subvolumes to predict polyp class, and model SEG was additionally trained with polyp segmentation masks. Diagnostic performance was validated in an independent external multicentre test sample. Predictions were analysed with the visualisation technique Grad-CAM++. RESULTS: The training set consisted of 107 colorectal polyps in 63 patients (mean age: 63 ± 8 years, 40 men) comprising 169 polyp segmentations. The external test set included 77 polyps in 59 patients comprising 118 polyp segmentations. Model SEG achieved a ROC-AUC of 0.83 and 80% sensitivity at 69% specificity for differentiating premalignant from benign polyps. Model noSEG yielded a ROC-AUC of 0.75, 80% sensitivity at 44% specificity, and an average Grad-CAM++ heatmap score of ≥ 0.25 in 90% of polyp tissue. CONCLUSIONS: In this proof-of-concept study, deep learning enabled the differentiation of premalignant from benign colorectal polyps detected with CT colonography and the visualisation of image regions important for predictions. The approach did not require polyp segmentation and thus has the potential to facilitate the identification of high-risk polyps as an automated second reader. KEY POINTS: • Non-invasive deep learning image analysis may differentiate premalignant from benign colorectal polyps found in CT colonography scans. • Deep learning autonomously learned to focus on polyp tissue for predictions without the need for prior polyp segmentation by experts. • Deep learning potentially improves the diagnostic accuracy of CT colonography in colorectal cancer screening by allowing for a more precise selection of patients who would benefit from endoscopic polypectomy, especially for patients with polyps of 6-9 mm size.

Machine Learning-based Differentiation of Benign and Premalignant Colorectal Polyps Detected with CT Colonography in an Asymptomatic Screening Population: A Proof-of-Concept Study.

Background CT colonography does not enable definite differentiation between benign and premalignant colorectal polyps. Purpose To perform machine learning-based differentiation of benign and premalignant colorectal polyps detected with CT colonography in an average-risk asymptomatic colorectal cancer screening sample with external validation using radiomics. Materials and Methods In this secondary analysis of a prospective trial, colorectal polyps of all size categories and morphologies were manually segmented on CT colonographic images and were classified as benign (hyperplastic polyp or regular mucosa) or premalignant (adenoma) according to the histopathologic reference standard. Quantitative image features characterizing shape (n = 14), gray level histogram statistics (n = 18), and image texture (n = 68) were extracted from segmentations after applying 22 image filters, resulting in 1906 feature-filter combinations. Based on these features, a random forest classification algorithm was trained to predict the individual polyp character. Diagnostic performance was validated in an external test set. Results The random forest model was fitted using a training set consisting of 107 colorectal polyps in 63 patients (mean age, 63 years ± 8 [standard deviation]; 40 men) comprising 169 segmentations on CT colonographic images. The external test set included 77 polyps in 59 patients comprising 118 segmentations. Random forest analysis yielded an area under the receiver operating characteristic curve of 0.91 (95% CI: 0.85, 0.96), a sensitivity of 82% (65 of 79) (95% CI: 74%, 91%), and a specificity of 85% (33 of 39) (95% CI: 72%, 95%) in the external test set. In two subgroup analyses of the external test set, the area under the receiver operating characteristic curve was 0.87 in the size category of 6-9 mm and 0.90 in the size category of 10 mm or larger. The most important image feature for decision making (relative importance of 3.7%) was quantifying first-order gray level histogram statistics. Conclusion In this proof-of-concept study, machine learning-based image analysis enabled noninvasive differentiation of benign and premalignant colorectal polyps with CT colonography. © RSNA, 2021 Online supplemental material is available for this article.

Pneumothorax detection in chest radiographs: optimizing artificial intelligence system for accuracy and confounding bias reduction using in-image annotations in algorithm training.

OBJECTIVES: Diagnostic accuracy of artificial intelligence (AI) pneumothorax (PTX) detection in chest radiographs (CXR) is limited by the noisy annotation quality of public training data and confounding thoracic tubes (TT). We hypothesize that in-image annotations of the dehiscent visceral pleura for algorithm training boosts algorithm's performance and suppresses confounders. METHODS: Our single-center evaluation cohort of 3062 supine CXRs includes 760 PTX-positive cases with radiological annotations of PTX size and inserted TTs. Three step-by-step improved algorithms (differing in algorithm architecture, training data from public datasets/clinical sites, and in-image annotations included in algorithm training) were characterized by area under the receiver operating characteristics (AUROC) in detailed subgroup analyses and referenced to the well-established "CheXNet" algorithm. RESULTS: Performances of established algorithms exclusively trained on publicly available data without in-image annotations are limited to AUROCs of 0.778 and strongly biased towards TTs that can completely eliminate algorithm's discriminative power in individual subgroups. Contrarily, our final "algorithm 2" which was trained on a lower number of images but additionally with in-image annotations of the dehiscent pleura achieved an overall AUROC of 0.877 for unilateral PTX detection with a significantly reduced TT-related confounding bias. CONCLUSIONS: We demonstrated strong limitations of an established PTX-detecting AI algorithm that can be significantly reduced by designing an AI system capable of learning to both classify and localize PTX. Our results are aimed at drawing attention to the necessity of high-quality in-image localization in training data to reduce the risks of unintentionally biasing the training process of pathology-detecting AI algorithms. KEY POINTS: • Established pneumothorax-detecting artificial intelligence algorithms trained on public training data are strongly limited and biased by confounding thoracic tubes. • We used high-quality in-image annotated training data to effectively boost algorithm performance and suppress the impact of confounding thoracic tubes. • Based on our results, we hypothesize that even hidden confounders might be effectively addressed by in-image annotations of pathology-related image features.

End-to-End Deep Learning Approach for Perfusion Data: A Proof-of-Concept Study to Classify Core Volume in Stroke CT.

(1) Background: CT perfusion (CTP) is used to quantify cerebral hypoperfusion in acute ischemic stroke. Conventional attenuation curve analysis is not standardized and might require input from expert users, hampering clinical application. This study aims to bypass conventional tracer-kinetic analysis with an end-to-end deep learning model to directly categorize patients by stroke core volume from raw, slice-reduced CTP data. (2) Methods: In this retrospective analysis, we included patients with acute ischemic stroke due to proximal occlusion of the anterior circulation who underwent CTP imaging. A novel convolutional neural network was implemented to extract spatial and temporal features from time-resolved imaging data. In a classification task, the network categorized patients into small or large core. In ten-fold cross-validation, the network was repeatedly trained, evaluated, and tested, using the area under the receiver operating characteristic curve (ROC-AUC). A final model was created in an ensemble approach and independently validated on an external dataset. (3) Results: 217 patients were included in the training cohort and 23 patients in the independent test cohort. Median core volume was 32.4 mL and was used as threshold value for the binary classification task. Model performance yielded a mean (SD) ROC-AUC of 0.72 (0.10) for the test folds. External independent validation resulted in an ensembled mean ROC-AUC of 0.61. (4) Conclusions: In this proof-of-concept study, the proposed end-to-end deep learning approach bypasses conventional perfusion analysis and allows to predict dichotomized infarction core volume solely from slice-reduced CTP images without underlying tracer kinetic assumptions. Further studies can easily extend to additional clinically relevant endpoints.

Impact of Confounding Thoracic Tubes and Pleural Dehiscence Extent on Artificial Intelligence Pneumothorax Detection in Chest Radiographs.

OBJECTIVES: We hypothesized that published performances of algorithms for artificial intelligence (AI) pneumothorax (PTX) detection in chest radiographs (CXRs) do not sufficiently consider the influence of PTX size and confounding effects caused by thoracic tubes (TTs). Therefore, we established a radiologically annotated benchmarking cohort (n = 6446) allowing for a detailed subgroup analysis. MATERIALS AND METHODS: We retrospectively identified 6434 supine CXRs, among them 1652 PTX-positive cases and 4782 PTX-negative cases. Supine CXRs were radiologically annotated for PTX size, PTX location, and inserted TTs. The diagnostic performances of 2 AI algorithms ("AI_CheXNet" [Rajpurkar et al], "AI_1.5" [Guendel et al]), both trained on publicly available datasets with labels obtained from automatic report interpretation, were quantified. The algorithms' discriminative power for PTX detection was quantified by the area under the receiver operating characteristics (AUROC), and significance analysis was based on the corresponding 95% confidence interval. A detailed subgroup analysis was performed to quantify the influence of PTX size and the confounding effects caused by inserted TTs. RESULTS: Algorithm performance was quantified as follows: overall performance with AUROCs of 0.704 (AI_1.5) / 0.765 (AI_CheXNet) for unilateral PTXs, AUROCs of 0.666 (AI_1.5) / 0.722 (AI_CheXNet) for unilateral PTXs smaller than 1 cm, and AUROCs of 0.735 (AI_1.5) / 0.818 (AI_CheXNet) for unilateral PTXs larger than 2 cm. Subgroup analysis identified TTs to be strong confounders that significantly influence algorithm performance: Discriminative power is completely eliminated by analyzing PTX-positive cases without TTs referenced to control PTX-negative cases with inserted TTs. Contrarily, AUROCs increased up to 0.875 (AI_CheXNet) for large PTX-positive cases with inserted TTs referenced to control cases without TTs. CONCLUSIONS: Our detailed subgroup analysis demonstrated that the performance of established AI algorithms for PTX detection trained on public datasets strongly depends on PTX size and is significantly biased by confounding image features, such as inserted TTS. Our established, clinically relevant and radiologically annotated benchmarking cohort might be of great benefit for ongoing algorithm development.

Experimental comparison of proton CT and dual energy x-ray CT for relative stopping power estimation in proton therapy.

Proton computed tomography (pCT) has been proposed as an alternative to x-ray computed tomography (CT) for acquiring relative to water stopping power (RSP) maps used for proton treatment planning dose calculations. In parallel, it has been shown that dual energy x-ray CT (DECT) improves RSP accuracy when compared to conventional single energy x-ray CT. This study aimed at directly comparing the RSP accuracy of both modalities using phantoms scanned at an advanced prototype pCT scanner and a state-of-the-art DECT scanner. Two phantoms containing 13 tissue-mimicking inserts of known RSP were scanned at the pCT phase II prototype and a latest generation dual-source DECT scanner (Siemens SOMATOM Definition FORCE). RSP accuracy was compared by mean absolute percent error (MAPE) over all inserts. A highly realistic Monte Carlo (MC) simulation was used to gain insight on pCT image artifacts which degraded MAPE. MAPE was 0.55% for pCT and 0.67% for DECT. The realistic MC simulation agreed well with pCT measurements ([Formula: see text]). Both simulation and experimental results showed ring artifacts in pCT images which degraded the MAPE compared to an ideal pCT simulation ([Formula: see text]). Using the realistic simulation, we could identify sources of artifacts, which are attributed to the interfaces in the five-stage plastic scintillator energy detector and calibration curve interpolation regions. Secondary artifacts stemming from the proton tracker geometry were also identified. The pCT prototype scanner outperformed a state-of-the-art DECT scanner in terms of RSP accuracy (MAPE) for plastic tissue mimicking inserts. Since artifacts tended to concentrate in the inserts, their mitigation may lead to further improvements in the reported pCT accuracy.