# Case of the day: one real patient a day, for doctors and medical students

From open medical datasets: a stethoscope recording, an ECG, an X-ray, a lung CT or a labour CTG trace. One question, then what was actually found. Today, the same case for everyone. Made in Romania for its resident doctors, from five research datasets gathered in Europe, the Americas and Asia.


## What Case of the day is

Every day, one real patient: a stethoscope recording, a 12-lead ECG, an X-ray, a lung CT or a CTG trace from labour. There is one closed question, answered in about a minute. Then the same material shows what was actually found: where the murmur was, what the cardiologist wrote, where the fracture is, where each radiologist drew, what pH the baby had at birth.
The case changes at midnight, Romanian time (UTC+2 in winter), and is the same for everyone. The grid of squares can be sent to colleagues without giving the answer away. A link to a case opens exactly that case.
It is practice, not an exam and not medical advice. Nothing here replaces examining a patient.

## How well do young doctors recognise heart sounds and murmurs?

A 1997 American study (Mangione and Nieman, JAMA) had 453 internal medicine and family practice residents, plus 88 medical students, listen to 12 cardiac events recorded from patients (extra heart sounds and murmurs) and pick the answer from several options. On average the residents recognised 20% of them. Scores rose little with year of training and were not significantly better than the students'.
No such measurement exists for Romania, the country this site was made for. And the score here does not compare with that 20%: there it was 12 different events in adults, with multiple options; here, at the stethoscope, the question is whether a child has a murmur, and a coin toss would get half the answers right.

## How is the right answer known?

Each kind of case has one question, checked against one label from the source. The kind of truth is written on every case, because it is not the same everywhere.

CaseQuestionChecked against

- Stethoscope: “Do you hear a murmur?” or “Would you refer this child?” — the annotation of the experts who listened to the recordings, or the paediatric cardiologist's verdict, made with the history, the examination and, where needed, an echo
- ECG: one choice out of four — the cardiologist's label, only on records validated by a person; not an angiogram
- X-ray: “Where is the fracture?” or “there is none” — the outline drawn by the annotators, with the help of two radiologists and an orthopaedic surgeon
- Lung CT: “Where is the nodule?” — how many of the four radiologists outlined it: agreement and disagreement
- CTG: “What pH was this baby born with?”: acidaemia or normal, read from the trace — the umbilical artery pH measured at birth: below 7.05 is acidaemia, 7.20 or above is normal; cases in between are left out. The pH cannot say what should have been done in labour, so the question is not about management

The cases were drawn at random, by rules fixed before any of them was seen, not picked for how they look. Uncertain labels (murmur “unknown”, ECGs not validated by a person) and recordings that were too noisy were set aside. Every case passes an automatic check: the licence, the identifier in the source, the label read from the original file, and the match between question and label.
The datasets come from groups selected for research (voluntary screening of children in north-east Brazil, hospitals in Bangladesh, the Czech Republic and the United States). How often a disease turns up here says nothing about any population.

## Where do the cases come from?

- Stethoscope: [CirCor DigiScope Phonocardiogram Dataset 1.0.3](https://physionet.org/content/circor-heart-sound/1.0.3/), PhysioNet, ODC-BY 1.0 licence. Recordings from children and teenagers in two screening campaigns in north-east Brazil, 2014 and 2015. Oliveira J. et al., IEEE J Biomed Health Inform 2022;26:2524–2535.
- ECG: [PTB-XL 1.0.3](https://physionet.org/content/ptb-xl/1.0.3/), PhysioNet, CC BY 4.0 licence. 21,799 ten-second ECGs recorded between 1989 and 1996 on Schiller machines, labelled by one or two cardiologists. Wagner P. et al., Sci Data 2020;7:154.
- X-ray: [FracAtlas](https://doi.org/10.6084/m9.figshare.22363012), figshare, CC BY 4.0 licence. 4,083 radiographs from three hospitals in Bangladesh, annotated for fractures. Abedeen I. et al., Sci Data 2023;10:521.
- Lung CT: [LIDC-IDRI](https://www.cancerimagingarchive.net/collection/lidc-idri/), via Imaging Data Commons, CC BY 3.0 licence. 1,018 chest CTs, each read by four thoracic radiologists, first separately, then seeing the others' marks. Armato SG III et al., Med Phys 2011;38:915–931.
- CTG: [CTU-UHB Intrapartum Cardiotocography Database 1.0.0](https://physionet.org/content/ctu-uhb-ctgdb/1.0.0/), PhysioNet, ODC-BY 1.0 licence. 552 labour traces from the University Hospital in Brno, Czech Republic, 2010–2012, with cord pH. Chudáček V. et al., BMC Pregnancy Childbirth 2014;14:16.
- The listening study: Mangione S., Nieman L.Z. [Cardiac auscultatory skills of internal medicine and family practice trainees](https://pubmed.ncbi.nlm.nih.gov/9286830/). JAMA 1997;278:717–722.

On the same data, the page [The baby's heart on the monitor](https://mariuscomper.uk/bebelusul-pe-monitor/en/) shows what nine obstetricians said about each Brno trace.
Other pages built on real medical images: [The brain seen by ultrasound](https://mariuscomper.uk/creierul-vazut-prin-ecograf/en/) and [Laparoscopic ultrasound on a real liver](https://mariuscomper.uk/ficatul-vazut-prin-ecograf/en/).