AI-based detection and classification of distal radius fractures using low-effort data labeling: evaluation of applicability and effect of training set size

Objectives

To evaluate the performance of a deep convolutional neural network (DCNN) in detecting and classifying distal radius fractures, metal, and cast on radiographs using labels based on radiology reports. The secondary aim was to evaluate the effect of the training set size on the algorithm’s performance.

Methods

A total of 15,775 frontal and lateral radiographs, corresponding radiology reports, and a ResNet18 DCNN were used. Fracture detection and classification models were developed per view and merged. Incrementally sized subsets served to evaluate effects of the training set size. Two musculoskeletal radiologists set the standard of reference on radiographs (test set A). A subset (B) was rated by three radiology residents. For a per-study-based comparison with the radiology residents, the results of the best models were merged. Statistics used were ROC and AUC, Youden’s J statistic (J), and Spearman’s correlation coefficient (ρ).

Results

The models’ AUC/J on (A) for metal and cast were 0.99/0.98 and 1.0/1.0. The models’ and residents’ AUC/J on (B) were similar on fracture (0.98/0.91; 0.98/0.92) and multiple fragments (0.85/0.58; 0.91/0.70). Training set size and AUC correlated on metal (ρ = 0.740), cast (ρ = 0.722), fracture (frontal ρ = 0.947, lateral ρ = 0.946), multiple fragments (frontal ρ = 0.856), and fragment displacement (frontal ρ = 0.595).

Conclusions

The models trained on a DCNN with report-based labels to detect distal radius fractures on radiographs are suitable to aid as a secondary reading tool; models for fracture classification are not ready for clinical use. Bigger training sets lead to better models in all categories except joint affection.

Key Points

• Detection of metal and cast on radiographs is excellent using AI and labels extracted from radiology reports.

• Automatic detection of distal radius fractures on radiographs is feasible and the performance approximates radiology residents.

• Automatic classification of the type of distal radius fracture varies in accuracy and is inferior for joint involvement and fragment displacement.

Total versus unspecific binding: standardization and optimization of receptor-ligand binding assays

Intra- and interassay imprecision was evaluated in 3 in vitro ligand binding assays using different types of filtering devices, glass fiber filters and variations in methodology. In the [3H]-spiroperidol test, the already low unspecific binding seemed to be dependent on the type of filters employed, whereas in the [3H]-ketanserin- and [3H]-GR 65630-binding tests, differences were within the range of the normal variabilities. However, in the latter test, which is problematic owing to the high unspecific binding of [3H]-GR 65630, it was found that although the percentage of specific binding was fairly constant on different sheets, large differences in the absolute amounts of total and unspecific binding were observed on consecutive sheets in the same experiment. Thus, it is critically important to filter samples for total and unspecific binding together on the same filter sheet for the calculation of specific binding. Under these precautions, highly reproducible results for IC50-values in the screening of potential 5HT3-receptor ligands were obtained in spite of using rat entorhinal cortex, a relatively large area with suboptimal 5HT3-receptor density. In contrast, when using rat area postrema, which is optimal with respect to receptor density, more than 10 times the number of rats is necessary for a competition experiment due to the small size of this brain part. Since IC50-values for both areas compare favorably, entorhinal cortex should be used for ethical reasons and to minimize costs.

High-performance liquid chromatographic system for the separation of dynorphin A (1-17) fragments and its application in enzymolysis studies with rat nerve terminal membranes

A high-performance liquid chromatographic method capable of separating a large number of C- and N-terminal degradation fragments of dynorphin A (1-17) (dyn 1-17) in 1 h has been developed. The system has been applied to study the metabolism profiles of various dyn 1-17-derived peptides following in vitro incubation with rat striatum and spinal cord nerve terminal membranes. In addition to the removal of the N-terminal amino acid Tyr, major sites of cleavage between the following amino acids could be established: Leu5-Arg6 in dyn 1-7 (formation of dyn 1-5); Arg6-Arg7 and Leu5-Arg6 in dyn 1-8 (formation of dyn 1-6 and dyn 1-5, respectively); Arg7-Ile8 in dyn 1-9 (formation of dyn 1-7) and Arg9-Pro10 in dyn 1-10 (formation of dyn 1-9). Studies with inhibitors of the enzymes involved show that dyn 1-5 is formed directly from dyn 1-8 via an endopeptidase insensitive to the angiotensin-converting enzyme inhibitor MK 422 acting on the scissile Leu5-Arg6 bond in dyn 1-8. The method circumvents the use of [3H]Tyr-labelled dynorphins, which have the inherent drawback that fragments lacking the N-terminal Tyr cannot be detected. Owing to the high resolution, also for the larger dynorphins dyn 1-14, dyn 1-15 and dyn 1-16, the chromatographic system should prove especially useful in the elucidation of the enzymolysis pattern of dyn 1-17. Furthermore, the method offers a way to evaluate simultaneously the selectivity of new enzyme inhibitors for several cleavage sites in the same assay.

Acute hypoproteinemic fluid overload: its determinants, distribution, and treatment with concentrated albumin and diuretics

We simulated the use of massive volumes of crystalloid fluids as a treatment of acute plasma loss in a standardized experimental model and studied the factors determining the retention or excretion of the resulting acute hypoproteinemic fluid overload, its distribution within the body, and its treatment with concentrated albumin and diuretics. In accordance with the classic Starling concept, the serum protein level, i.e. the serum colloid osmotic pressure, determined the excretion/retention ratio of a given water and sodium load. Of the total fluid retention, fat and muscle each accommodated 25%, whereas the skin, which contributes only 7% to the total body weight, accounted for 37% and increased its volume by roughly one third. Concentrated albumin promoted fluid excretion in direct proportion to the achieved increment of the serum protein level and abolished the edema of fat, muscle and skin. Furosemide was virtually ineffective. The implications of these results for the 'adult respiratory distress syndrome' and disturbed wound healing are discussed and related to the concept of a critical threshold of the serum protein level.