Performance of Progressive Generations of GPT on an Exam Designed for Certifying Physicians as Certified Clinical Densitometrists

BACKGROUND

Artificial intelligence (AI) large language models (LLMs) such as ChatGPT have demonstrated the ability to pass standardized exams. These models are not trained for a specific task, but instead trained to predict sequences of text from large corpora of documents sourced from the internet. It has been shown that even models trained on this general task can pass exams in a variety of domain-specific fields, including the United States Medical Licensing Examination. We asked if large language models would perform as well on a much narrower subdomain tests designed for medical specialists. Furthermore, we wanted to better understand how progressive generations of GPT (generative pre-trained transformer) models may be evolving in the completeness and sophistication of their responses even while generational training remains general. In this study, we evaluated the performance of two versions of GPT (GPT 3 and 4) on their ability to pass the certification exam given to physicians to work as osteoporosis specialists and become a certified clinical densitometrists. The CCD exam has a possible score range of 150 to 400. To pass, you need a score of 300.

METHODS

A 100-question multiple-choice practice exam was obtained from a 3rd party exam preparation website that mimics the accredited certification tests given by the ISCD (International Society for Clinical Densitometry). The exam was administered to two versions of GPT, the free version (GPT Playground) and ChatGPT+, which are based on GPT-3 and GPT-4, respectively (OpenAI, San Francisco, CA). The systems were prompted with the exam questions verbatim. If the response was purely textual and did not specify which of the multiple-choice answers to select, the authors matched the text to the closest answer. Each exam was graded and an estimated ISCD score was provided from the exam website. In addition, each response was evaluated by a rheumatologist CCD and ranked for accuracy using a 5-level scale. The two GPT versions were compared in terms of response accuracy and length.

RESULTS

The average response length was 11.6 ±19 words for GPT-3 and 50.0±43.6 words for GPT-4. GPT-3 answered 62 questions correctly resulting in a failing ISCD score of 289. However, GPT-4 answered 82 questions correctly with a passing score of 342. GPT-3 scored highest on the "Overview of Low Bone Mass and Osteoporosis" category (72 % correct) while GPT-4 scored well above 80 % accuracy on all categories except "Imaging Technology in Bone Health" (65 % correct). Regarding subjective accuracy, GPT-3 answered 23 questions with nonsensical or totally wrong responses while GPT-4 had no responses in that category.

CONCLUSION

If this had been an actual certification exam, GPT-4 would now have a CCD suffix to its name even after being trained using general internet knowledge. Clearly, more goes into physician training than can be captured in this exam. However, GPT algorithms may prove to be valuable physician aids in the diagnoses and monitoring of osteoporosis and other diseases.

[Translated article] Influence of lateralized reverse shoulder prosthesis design on tuberosity union in proximal humerus fractures

INTRODUCTION

Proximal humerus fractures are the third most frequent type of fracture in elderly patients. Nowadays, surgical treatment is indicated one third of the time, being the reverse shoulder prosthesis an option especially in complex comminuted patterns. In this study we analyzed the effects of a lateralized reverse prosthesis in tuberosity union and its relationship with the functional results.

MATERIAL AND METHODS

Retrospective case study of patients with proximal humerus fractures treated with a lateralized design reverse shoulder prosthesis with one-year minimum follow-up. Tuberosity nonunion was defined as a radiological concept: absence of tuberosity, distance >1cm from the tuberosity fragment to the humeral shaft or tuberosity above the humeral tray. Subgroup analysis was performed, group 1 (n=16) tuberosity union vs. group 2 (n=19) tuberosity nonunion. Groups were compared with the following functional scores: Constant, American Shoulder and Elbow Surgeons and Subjective Shoulder Value.

RESULTS

A total of 35 patients were included in this study with a median age of 72.65 years. Postoperative radiographic analysis at one year after surgery revealed a tuberosity nonunion rate of 54%. Subgroup analysis revealed no statistically significant differences in terms of range of motion or functional scores. However, there were differences regarding the Patte sign (p=0.03) which was positive in a larger proportion of patients in the group with tuberosity nonunion.

CONCLUSION

Even though there was a large percentage of tuberosity nonunion with the use of a lateralized prosthesis design, patients obtained good results in a similar manner to those found in the union group in terms of range of motion, scores, and patient satisfaction.

Abstract P3-04-05: Artificial Intelligence Detects, Classifies, and Describes Lesions in Clinical Breast Ultrasound Images

Purpose Many low-middle income countries (LMIC) suffer from chronic shortages of resources that inhibit the implementation of effective breast cancer screening programs. Advanced breast cancer rates in the U.S. Affiliated Pacific Islands substantially exceed that of the United States. We propose the use of portable breast ultrasound coupled with artificial intelligence (AI) algorithms to assist non-radiologist field personnel in real-time field lesion detection, classification, and biopsy, as well as determination of breast density for risk assessment. In this study, we examine the ability of an AI algorithm to detect and describe breast cancer lesions in clinically-acquired breast ultrasound images in 40,000 women participating in a Hawaii screening program. Materials and Methods The Hawaii and Pacific Islands Mammography Registry (HIPIMR) collects breast health questionnaires and breast imaging (mammography, ultrasound, and MRI) from participating centers in Hawaii and the Pacific and links this information to the Hawaii Tumor Registry for cancer findings. From the women with either screening or diagnostic B-mode breast ultrasound exams, we selected 3 negative cases (no cancer) for every positive case matched by age, excluding Doppler and elastography images. The blinded images were read by the study radiologist to delineate all lesions and describe in terms of the BI-RADS lexicon. These images were split by woman into training (70%), validation and hyperparameter selection (20%) and testing (20%) subsets. An AI model was fine-tuned for lesion and BI-RADS category classification from a Detectron2 framework [1] pre-trained on the COCO Instance Segmentation Dataset [2]. Model performance was evaluated by computation of precision and sensitivity percentages, as well as Area under the Receiver Operator Curve (AUROC). Detections were considered positive if they overlapped a ground truth lesion delineation by at least 50% (Intersection over Union = 0.5), and a maximum of 4 detections were generated for each image. Timing experiments were run on a GPU-enabled (Nvidia Tesla V100) machine on unbatched images. Results Over the 10-year observation period, we identified 5,214 women with US images meeting our criterion. Of these, 111 were diagnosed with malignant breast cancer and 333 were selected as non-cases for a total of 444 women. These 444 women had a total of 4,623 ultrasound images with 2,028 benign and 1,431 malignant lesions identified by the study radiologist. For cancerous lesions, the AI algorithm had 8% precision at a sensitivity of 90% on the testing set. For benign lesions, a sensitivity of 90% resulted in 5% precision on the testing set. The AUROC for bounding box detections of cancerous lesions was 0.90. The AUROC for bounding box detections of benign lesions was 0.87. The model made predictions at a rate of 25 frames/second time (38.7 milliseconds per image). Conclusion Detection, segmentation, and cancer classification of breast lesions are possible in clinically-acquired ultrasound images using AI. Based on our timing experiments, the model is capable of detecting and classifying lesions in real-time during ultrasound capture. Model performance is expected to improve as more data becomes available for training. Future work would involve further fine-tuning of the model on portable breast ultrasound images and increasing model evaluation speed in order to assess utility in low-resource populations [1] Wu Y, Kirillov A, Massa F, Lo W-Y, Girshick R. Detectron2. https://github.com/facebookresearch/detectron2. [2] Lin T-Y, Maire M, Belongie S, et al. Microsoft COCO: Common Objects in Context. Computer Vision – ECCV 2014. Springer International Publishing; 2014:740-755.

Citation Format: Arianna Bunnell, Dustin Valdez, Thomas Wolfgruber, Aleen Altamirano, Brenda Hernandez, Peter Sadowski, John Shepherd. Artificial Intelligence Detects, Classifies, and Describes Lesions in Clinical Breast Ultrasound Images [abstract]. In: Proceedings of the 2022 San Antonio Breast Cancer Symposium; 2022 Dec 6-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2023;83(5 Suppl):Abstract nr P3-04-05.

Abstract P3-03-02: Can artificial intelligence derived ultrasound breast density provide comparable breast cancer risk estimates to density derived from mammograms

Can artificial intelligence derived ultrasound breast density provide comparable breast cancer risk estimates to density derived from mammograms Dustin Valdez12, Arianna Bunnell2, Thomas Wolfgruber1, Aleen V. Altamirano3, Brandon Quon1, Gertraud Maskarinec1, Peter Sadowski2, John A. Shepherd1 1 University of Hawaii Cancer Center, Honolulu, HI 2 University of Hawaii at Manoa, Honolulu, HI 3 Instituto Radiodiagnóstico, Managua, Nicaragua Background: Breast cancer is the second leading cause of cancer-related death among women in Hawaii and the Pacific. However, while there are programs like the Breast and Cervical Cancer Early Detection Program (BCCEDP) implemented throughout the Pacific, the lack of access to mammography screening and low screening participation rates contributes to very high advanced breast cancer rates in most cases over 50%. Portable breast ultrasound is a promising screening technology for low resource areas. However, without mammography, mammographic density is not available for risk modeling to determine who should participate in screening programs or at what frequency. In this study, we ask if breast ultrasound (US) images can be used to derive an equivalent mammographic density for risk modeling. We utilized artificial intelligence to derive breast density from diagnostic ultrasound images and compared to BI-RADS mammographic density in an established breast cancer risk model1. Methods: We selected women with negative screening visit who either later developed cancer (positives) or did not (negatives) over a 10-year period. Temporally-matched negative mammographic and ultrasound images, cancer outcome status and cancer risk information were sourced from the Hawaii and Pacific Islands Mammography Registry. US images had to have occurred within a year of the mammogram. BI-RADS mammographic density was derived using an existing deep neural network model2. Mammographic density was estimated from US images by training a deep-learning convolutional neural network model. A hold out set of images (Test set of 20% of the total) was used to compare 10-year breast cancer risk using the Tyrer-Cuzick (TC) risk model1 when calculated using breast density from either mammograms or US. The AUC values, confidence intervals, ROC plots and Pearson correlation were calculated and compared. Results: Over the 10-year study period, 1337 had matched mammograms and US images and 65 went on to develop breast cancer. Using the test set, the Pearson’s correlation between breast density from mammography and US was 0.31 (moderate correlation). There were no covariates found to improve this association. The AUC for TC 10-year personal risk was higher when breast density from mammograms was used 0.71 (95% CI=0.57-0.86) versus US images 0.65 (95% CI=0.53-0.76). Conclusion: Overall breast cancer risk was similar when breast density was derived from either mammograms or US. The performance of our US breast density model is expected to improve further when more US training data becomes available. Breast cancer screening programs exclusively using US imaging may be able to provide equivalent risk modeling to clinics using mammography. 1.Tyrer J, Duffy SW, Cuzick J (2004). A breast cancer prediction model incorporating familial and personal risk factors. Stat Med. 2004 Apr 15;23(7):1111-30. doi: 10.1002/sim.1668. Erratum in: Stat Med. 2005 Jan 15;24(1):156. PMID: 15057881. 2. Wu, N., K. J. Geras, Y. Shen, J. Su, S. G. Kim, E. Kim, S. Wolfson, L. Moy and K. Cho (2018). Breast Density Classification with Deep Convolutional Neural Networks. 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), IEEE.

Citation Format: Dustin Valdez, Arianna Bunnell, Thomas Wolfgruber, Aleen Altamirano, Brandon Quon, Gertraud Maskarinec, Peter Sadowski, John Shepherd. Can artificial intelligence derived ultrasound breast density provide comparable breast cancer risk estimates to density derived from mammograms [abstract]. In: Proceedings of the 2022 San Antonio Breast Cancer Symposium; 2022 Dec 6-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2023;83(5 Suppl):Abstract nr P3-03-02.

Technical note: Low clinical efficacy, but good acceptability of a point-of-care electronic palpation device for breast cancer screening for a lower middle-income environment

BACKGROUND

Late-stage breast cancer rates in the Pacific where mammography services are limited are exceedingly high: Marshall Islands (61%), Palau (94%), and Samoa (79%). Due to the limited medical resources in these areas an alternative accessible technology is needed. The iBreast Exam (iBE) is a point-of-care electronic palpitation device that has a reported sensitivity of 86%. However, little is known about the performance and acceptability of this device for women in the Pacific.

METHODS

A total of 39 women (ages 42-73 years) were recruited in Guam with 19 women having a mammogram requiring biopsy (Breast Imaging-Reporting and Data System [BI-RADS] category 4 or above) and 20 women with a negative screening mammogram before the study visit. Participants received an iBE exam and completed a 26-item breast health questionnaire to evaluate the iBE. Furthermore, the performance characteristics of the iBE were tested using gelatin breast phantoms in terms of tumor size, tumor depth, and overall breast stiffness.

RESULTS

The iBE had a sensitivity of 20% (two true positives to eight false negatives) and specificity of 92% (24 false positives to 278 true negatives) when analyzed based on the location of the tumor by quadrant. The iBE also had generally poor agreement according to a Cohen's kappa value of 0.068. The phantom experiments showed that the iBE can detect tumors as deep as 2.5 cm, but only if the lesion is greater than 8 mm in diameter. However, the iBE did demonstrate acceptability; 67% of the women reported that they had high trust in iBE as an early detection device.

CONCLUSIONS

The iBE had generally poor sensitivity and specificity when tested in a clinical setting which does not allow its use as a screening tool.

IMPACT

This study demonstrates the need for an alternative screening method other than electronic palpation for lower-middle-income areas.

Abstract P3-01-13: Comparing portable and clinical ultrasound systems using 3D printed breast phantom inserts

Background: Late-stage breast cancer rates in the Pacific where mammography services are limited are exceedingly high. Therefore, alternative accessible breast cancer screening technologies such as portable ultrasound is needed. However, little is known about the performance of portable ultrasound when compared to clinical ultrasound for use in breast cancer screening. By utilizing 3D printing technology, it is possible to design breast phantom inserts to replicate various types of lesions. In this study, we utilized 3D printed breast phantom inserts to compare portable and clinical ultrasound lesion detection performance. Methods: Four different breast inserts were designed using FreeCAD (version 0.19) to replicate different lesion detection properties. The first insert compares lesion shape, the second insert investigates depth and size, the third insert looks at fiber diameter, and the fourth insert looks at clusters. The four inserts were printed using a photopolymer resign (Formlabs Inc Rigid resign, Somerville, MA, USA) and then placed in a gelatin-based breast phantom designed for ultrasound use. Using the portable ultrasound (GE Vscan Extend) and clinical ultrasound (Philips EPIQ 5G), various images were captured of identical angle and orientation for both devices. The number of lesions visualized were counted and presented as a percentage of lesions detected. Results: The portable ultrasound had a 100% lesion detection rate for breast insert 1, 90.3% for breast insert 2, 70% for breast insert 3 and 55.8% for breast insert 4. Clinical ultrasound had 100% lesion detection rate for breast insert 1, 93.1% for breast insert 2, 76.6% for breast insert 3, and 99% for breast insert 4. Conclusion: Portable ultrasound shows comparable lesion detection capabilities to clinical ultrasound in 3 of the 4 breast phantom insert tests. Portable ultrasound may have potential as a capable accessible breast cancer screening device in areas without mammography.

Citation Format: Dustin Valdez, Jami Fukui, Thomas Wolfgruber, Lambert Leong, Gertraud Maskarinec, John Shepherd. Comparing portable and clinical ultrasound systems using 3D printed breast phantom inserts [abstract]. In: Proceedings of the 2021 San Antonio Breast Cancer Symposium; 2021 Dec 7-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2022;82(4 Suppl):Abstract nr P3-01-13.

Determining Acceptability of a Text-message Based Intervention to Prevent Excessive Gestational Weight Gain in Low-income Women in Hawai'i WIC (P04-029-19)

Objectives

The purpose of this study was to determine the acceptability of messages in a text message-based nutrition intervention for the prevention of excessive gestational weight gain in low-income women in the Women, Infants, and Children (WIC) program in Hawai‘i.

Methods

Low-income pregnant women (n = 100) in Hawaii participated in a four-month text message-based nutrition intervention program. Participants in the intervention arm received 18 text messages (1/week) that focused on reinforcing WIC's nutritional recommendations for pregnant women. Participants who completed the intervention were interviewed at their respective WIC clinic about their experiences with the messages. The interviews were transcribed and content analysis to identify the prevailing themes and concepts was performed with NVivo (version 12, GSR International, Inc, Burlington, MA).

Results

Participants responded to questions in four content areas: most useful messages for staying healthy, least useful messages for staying healthy, messages that affected eating and exercise habits, and the experience of receiving messages. The most useful messages related to healthy food substitutions. The least helpful messages were those participants felt they were unable to perform, such as eating sardines for omega-3 fatty acids. Participants cited that messages relating to healthy food substitutions as having the most impact on eating behavior. Most participants also felt that increasing the number of messages would have been helpful.

Conclusions

Results suggest that providing information relating to healthy food substitutions as well as providing a high frequency of messages could be beneficial in the development of mobile health programs for low-income pregnant women in controlling gestational weight gain.

Funding Sources

Mountain West Clinical Translational Research Infrastructure Network under a grant from National Institute of General Medicine Sciences of the National Institute of Health.

Hydration and protein folding in water and in reverse micelles: compressibility and volume changes

The partial specific volume and adiabatic compressibility of proteins reflect the hydration properties of the solvent-exposed protein surface, as well as changes in conformational states. Reverse micelles, or water-in-oil microemulsions, are protein-sized, optically-clear microassemblies in which hydration can be experimentally controlled. We explore, by densimetry and ultrasound velocimetry, three basic proteins: cytochrome c, lysozyme, and myelin basic protein in reverse micelles made of sodium bis (2-ethylhexyl) sulfosuccinate, water, and isooctane and in aqueous solvents. For comparison, we use beta-lactoglobulin (pI = 5.1) as a reference protein. We examine the partial specific volume and adiabatic compressibility of the proteins at increasing levels of micellar hydration. For the lowest water content compatible with complete solubilization, all proteins display their highest compressibility values, independent of their amino acid sequence and charge. These values lie within the range of empirical intrinsic protein compressibility estimates. In addition, we obtain volumetric data for the transition of myelin basic protein from its initially unfolded state in water free of denaturants, to a folded, compact conformation within the water-controlled microenvironment of reverse micelles. These results disclose yet another aspect of the protein structural properties observed in membrane-mimetic molecular assemblies.

Induction of cortical swallowing activity by transcranial magnetic stimulation in the anaesthetized cat

Transcranial magnetic stimulation (TMS) over human fronto-central areas of scalp can activate short latency responses in the muscles of the face, pharynx and oesophagus. However, the physiological relationship between this early activity and the swallowing activity programmed by the brainstem central pattern generator (CPG) remains unclear. The aim of this study was to explore the relationship between TMS-induced early muscle and late swallowing activities in the feline model. Twelve adult cats were studied under light anaesthesia. Mylohyoid and oesophageal EMG, together with pharyngeal, upper oesophageal sphincter (UOS) and upper oesophageal manometry, were recorded to single-pulse TMS of cat cortex. TMS at low stimulation intensities evoked consistent short latency EMG responses in the mylohyoid and oesophageal muscles (6.1 +/- 1.2 ms and 12.7 +/- 0.7 ms, respectively), and early contractile activity in the UOS (latency 31.8 +/- 3.6 ms). By contrast, TMS at high intensities induced swallowing activity as indicated by mylohyoid EMG, and UOS relaxation (latencies 1.1 +/- 0.4 s and 0.8 +/- 0.1 s, respectively). Both the early muscle and late swallowing activities were intensity-dependent, increasing stimulus strength producing a reduction in latency and greater number of swallows. The characteristics of the early response suggest an oligosynaptic projection from cortex to swallowing muscles. The induction of swallows at high intensities suggests a requisite for greater recruitment of cortical motoneurones, or associated swallowing regions.

Adiabatic compressibility of AOT

The self-assembly of amphiphilic molecules into supramolecular aggregates involves a number of complex phenomena and forces. Recent developments of highly sensitive, densimetric and acoustic methods on small volume samples have provided novel sensitive probes to explore the physical properties of these complex fluids. We have investigated, by high precision densimetry and ultrasound velocimetry, reverse micelles of [sodium bis(2-ethylhexyl)sulfosuccinate] in oil (isooctane and decane), at increasing water concentration and at variable micellar volume fractions. The size of these spherical micelles has been determined by small angle x-ray scattering. Using these results, in the framework of the effective medium theory, we have developed a simple model of micellar compressibility, allowing the calculation of physical parameters (aggregation number, volume, and compressibility) of the surfactant monomolecular film as well as that of the micellar waters. In particular, we show that the central aqueous core designated as "free" water, located at a distance from the oil-water interacting interface, is twice as compressible as "bulk" water. One notable feature of this work is the influence of the nature of the oil on the above parameters.

Effect of an o-raffinose cross-linked haemoglobin product on oesophageal and lower oesophageal sphincter motor function

The present experiments evaluate the effects on oesophageal motility of an o-raffinose cross-linked haemoglobin-based oxygen carrier (HBOC) purified from outdated donated human blood cells (HemolinkTM), with attention to dose-response (0.6-2.4 g kg-1), oxygenation status and low molecular weight components (4.4-36.4% 64 kDa or less). In ketamine-anaesthetized cats, lower oesophageal sphincter (LES) function and oesophageal peristalsis were monitored 0.5 h before, during and up to 3.5 h after HBOC infusion, and in some cats at 24 h. (1) All products significantly inhibited LES relaxation and increased peristaltic velocity in the distal smooth muscle oesophagus, without consistently altering resting LES pressure. (2) Effects on peristaltic velocity reached a maximum at the smallest dose, whereas the effects on LES relaxation had a maximum effect at 1.2 g kg-1. (3) Effects were not significantly altered by the haemoglobin oxygenation status or presence of low molecular weight components. (4) Repetitive oesophageal contractions occurred. In the cat, an o-raffinose cross-linked human haemoglobin product produces changes in oesophageal body and LES function, which are independent of the HBOC oxygenation status and composition of the low molecular weight components tested. Changes may persist for at least 24 h. These motility changes are likely due to scavenging of nitric oxide by the haemoglobin.

Effects of nitric oxide synthase blockade on esophageal peristalsis and the lower esophageal sphincter in the cat

The present study explores the role of nitric oxide (NO) in control of esophageal peristalsis and lower esophageal sphincter (LES) function in the cat. Studies were performed on 20 ketamine-anesthetized cats with manometric recording at the LES, 0, 2, 4, and 6 cm above the LES (smooth muscle section), and 12 and (or) 14 cm above the LES (striated muscle section). L-Ng-Nitro-arginine (L-NNA, 10(-6)-10(-4) mol/kg) was given intravenously, and the effects on swallow-induced esophageal peristalsis were assessed. (i) L-NNA increased the velocity of swallow-induced peristalsis in the smooth muscle esophagus; the effect was dose dependent, more prominent distally, and completely reversed by L-arginine (10(-3) mol/kg). (ii) L-NNA decreased the amplitude of peristaltic contraction in the very distal esophagus; the decrease also was dose dependent but not returned to normal by L-arginine. (iii) L-NNA inhibited LES relaxation (reversed by L-arginine) and decreased the LES "after-contraction" amplitude (unaffected by L-arginine). (iv) L-NNA was associated with the appearance of repetitive contractions. Basal LES tone was unaffected by L-NNA. In conclusion, NO is an important mediator for the timing of peristalsis in the distal smooth muscle esophagus and for LES relaxation in the cat, a species whose contraction amplitude is largely determined by cholinergic excitation. The role of NO in controlling esophageal body and LES contraction amplitude, and in preventing repetitive contractions, requires further study.

Chloroflexus-like organisms from marine and hypersaline environments: Distribution and diversity

We report the presence of a diverse number ofChloroflexus-like organisms in intertidal marine and submerged hypersaline microbial mats using light, infrared fluorescence, and electron microscopy. The intertidal organisms appear morphologically very similar to thermophilicC. aurantiacus while the 2 hypersaline strains are larger and have a more complex ultrastructure composed of chlorosome-bearing internal membranes that appear to arise as invaginations of the cell membrane. By comparing spectroradiometry of microbial mat layers with microscopic observations, we have confirmed that theChloroflexus-like organisms are major constituents of the hypersaline microbial mat communities. In situ studies on mat layers dominated byChloroflexus-like organisms showed that sulfide-dependent photoautotrophic activity sustained by near infrared radiation prevailed. Autoradiographic analyses revealed that autotrophy was sustained in the filaments by 750 nm radiation. Three morphologically distinct strains are now maintained in mixed culture. One of these appears to be growing photoautotrophically.

Efficiency of thymidine incorporation in Escherichia coli B-r as a function of growth rate

Short-term labeling with radioactive thymidine led to inaccurate estimates of deoxyribonucleic acid synthesis at some growth rates in steady-state cultures of Escherichia coli B/r. Estimates were corrected in chemostat cultures by adding adenosine, a known inhibitor of thymidine phosphorylase.