How much will breathing capacity change after part of a lung is removed? A team in Taiwan measured how the operated lung actually changed shape, then compared that change with breathing tests. To read their result, put the scan and the breathing test on a timeline. The most informative imaging measurement needed a scan taken after the operation.
- There were 439 patients in the imaging study. Only 63 had the paired breathing tests used to evaluate the model.
- In an internal leave-one-out comparison, the CT-based approach had a mean absolute error of 9.1 percentage points in the change in forced vital capacity. The recalibrated segment-count method averaged 9.7 points.
- The 95% limits of agreement extended from −22.1 to +22.4 percentage points. A small average signed error can coexist with substantial differences for individual patients.
- Paper date
- 21 November 2026
- Setting
- National Taiwan University Hospital
- Imaging cohort
- 439 patients
- Paired breathing tests
- 63 patients, with follow-up at 3–6 months
Original publication: 21 Nov 2026 · The date above refers to this brief.
Average magnitude of the error in held-out patients
percentage points of FVC change
Place the inputs and outcome on a timeline
Before the operation
An initial CT and breathing test establish baseline measurements.
After the operation
A second CT captures the observed change on the operated side.
Evaluate the comparison
Repeat breathing tests supply the outcome in the 63-person subgroup.
When was the key measurement available?
The team used CT before and after surgery to estimate the volume of the operated lung. Because the depth of a breath can differ between scans, they used the opposite lung as a reference before calculating the change. The resulting measure captures the anatomy that remained and how it expanded after surgery. That makes it informative about recovery already under way. A model used the day before an operation would have to estimate that change from information available by then.
Source 1 ↗Where the sample gets smaller
All 439 people had usable imaging, but the breathing comparison required two spirometry assessments. Forced vital capacity, or FVC, measures the volume of air a person can forcibly breathe out after filling their lungs. Only 63 people had an eligible result both before surgery and three to six months afterwards. That is 14.4% of the imaging cohort. Surgeons ordered the follow-up tests for clinical reasons. People sent for testing may differ from those who had scans alone.
Source 1 ↗Two ways to average an error
The investigators repeatedly fitted the model on 62 patients and evaluated it on the one left out. Across the 63 predictions its mean absolute error was 9.1 percentage points of FVC change. The recalibrated anatomical comparator averaged 9.7 points, a difference of 0.6 points using the rounded results. The new approach also had a signed mean error of just +0.1 point. To see why the figures differ, imagine one estimate missing high by ten points and another missing low by ten. Their signed errors cancel. Their average absolute error is still ten points. The reported agreement limits show the spread that the signed average hides.
Source 1 ↗How an actual preoperative forecast would differ
An earlier study by Wu and colleagues used preoperative scans and breathing tests in 38 patients. Comparing its predicted and observed postoperative FVC gave a correlation of 0.86. A correlation describes how two sets of values track one another. An error in percentage points answers how far an estimate landed from its target. A useful next comparison would collect predictions before surgery, record outcomes consistently and report the errors in a separate group.
Source 1 ↗Source 2 ↗Separate average bias, error size and the timing of the scan
Publisher full text. Checked the sampling and timing of spirometry, formula for the imaging measure, cross-validation procedure, Figure 3 error estimates and limits of agreement. No patient-level records were examined.
01When did the model have its key CT input?
- What was observed
- The measure compared preoperative and postoperative volumes on the operated side, using the opposite lung as a reference.
- Where the conclusion stops
- It is observable only once the second scan exists.
Source 1 · Methods 2.3, contralateral normalization
02Did the small mean signed error imply tight agreement?
- What was observed
- Mean signed error was +0.1 percentage point, with 95% agreement limits from −22.1 to +22.4.
- Where the conclusion stops
- The signed average describes bias and can conceal opposite individual errors.
Source 1 · Results, out-of-fold agreement and Figure 3
Numbers you can inspect
| Measure | Value & unit | Origin & method |
|---|---|---|
| Imaging cohort | 439 patients | Reported Enrolled in the CT comparison. Source 1 · Methods 2.1 |
| Functional analysis | 63 patients | Reported Paired preoperative and postoperative pulmonary function tests. Source 1 · Methods 2.1 |
| Share of imaging cohort with paired tests | 14.4 percent | Calculated 63 divided by 439 times 100, rounded to one decimal. Source 1 · Methods 2.1 |
| Held-out CT-based mean absolute error | 0.091 fractional-change units | Reported Pooled held-out errors from the nested procedure. Multiply by 100 for 9.1 percentage points. Source 1 · Results, Figure 3B |
| Held-out recalibrated comparator error | 0.097 fractional-change units | Reported Segment-count comparator refitted inside each training fold. Multiply by 100 for 9.7 percentage points. Source 1 · Results, Figure 3B |
| Difference in absolute errors | 0.6 percentage points | Calculated Subtract 0.091 from 0.097 and multiply by 100, using the rounded published estimates. Source 1 · Results, Figure 3B |
| Mean signed error | 0.1 percentage point | Reported Mean out-of-fold difference in the agreement analysis. Source 1 · Results, Bland–Altman analysis |
| Lower 95% limit of agreement | -22.1 percentage points | Reported Lower estimated agreement limit. Source 1 · Results, Bland–Altman analysis |
| Upper 95% limit of agreement | 22.4 percentage points | Reported Upper estimated agreement limit. Source 1 · Results, Bland–Altman analysis |
| Span between agreement limits | 44.5 percentage points | Calculated Subtract −22.1 from +22.4. This is the width of the limits. Source 1 · Results, Bland–Altman analysis |
| Preoperative inputs in the older study | 38 patients | Reported Separate study of CT and spirometry obtained before surgery. Source 2 · Wu et al., published abstract |
| Correlation with postoperative FVC in the older study | 0.86 correlation | Reported Comparison of predicted and measured FVC in the earlier study. Source 2 · Wu et al., published abstract |
Compare the actual experiments
These studies answer different questions. Read the unit and endpoint before comparing results.
| Study | Unit & setting | Readout | Interpretation boundary |
|---|---|---|---|
| Earlier study using preoperative quantitative CT Source 2 · Wu et al., abstract, Methods and Results | 38 patients with preoperative CT and spirometry | Correlation of predicted and measured postoperative FVC | Correlation and absolute prediction error are different measures. |
| Postoperative volume and paired breathing tests Source 1 · Figure 3 and agreement analysis | 63 participants with follow-up spirometry | Out-of-fold signed and absolute errors | Mean signed error can be small even when individual estimates vary substantially. |
The export includes claims, available numbers, methods and source locations. It contains our reading notes and published summaries; it is not raw participant data or an independent reanalysis.
Evidence update · 22 Nov 2026
First publication. Any calculations in this brief use reported figures. The reading notes identify possible follow-up questions. AI source check; no human editorial or clinical review.
Read the error figures in their own units
| Measure | Reported or converted value | Interpretation |
|---|---|---|
| CT model, held-out mean absolute error | 9.1 percentage points | Average magnitude of the difference in FVC change. |
| Recalibrated segment count, held-out mean absolute error | 9.7 percentage points | Same outcome and held-out comparison. |
| CT model, mean signed error | +0.1 percentage point | Positive and negative errors can cancel. |
| CT model, 95% limits of agreement | −22.1 to +22.4 percentage points | Estimated spread of the individual differences. |
Figure 3 and out-of-fold results. The two absolute errors use the same scale. A 0.091 error on the proportional-change scale equals 9.1 percentage points.
Your questions, answered
Does this CT model estimate breathing capacity before surgery?
Its main imaging variable requires the observed change on a postoperative CT. A tool used before surgery would need a way to anticipate that value and its own validation.
Why do some figures say 439 patients and others say 63?
Imaging was available in the larger group. The smaller group also had eligible preoperative and postoperative breathing tests.
How can mean error be just +0.1 point when typical misses are larger?
A signed average allows high and low errors to cancel. Mean absolute error keeps their magnitudes. Here it was 9.1 points for the CT-based approach.
Can the correlation in the older study be compared directly with 9.1 points?
The older figure measures association. The newer figure measures distance from observed FVC change on a particular scale. They answer different questions.
Limits of this interpretation
- Follow-up spirometry was clinically selected, and only 63 patients entered the functional analysis.
- Leaving patients out one at a time provides internal testing within the same site.
- The main CT measure requires postoperative information. The reported limits of agreement are estimates in this sample.
Sources & transparency
- Chang et al. (2026) — Computed Tomography-Quantified Pre-to-Postoperative Operated-Side Volume Change Is Associated with Forced Vital Capacity Decline After Video-Assisted Thoracoscopic Anatomical Pulmonary Resection
Publisher full text. Checked the sampling and timing of spirometry, formula for the imaging measure, cross-validation procedure, Figure 3 error estimates and limits of agreement. No patient-level records were examined. · Accessed 22 Nov 2026
DOI: 10.3390/diagnostics16233917 - Wu et al. (1994) — Use of quantitative CT to predict postoperative lung function in patients with lung cancer
Publisher abstract of the separate 38-patient study. Checked timing of the input scans and spirometry and the reported correlation for FVC. · Accessed 22 Nov 2026
DOI: 10.1148/radiology.191.1.8134584
Prepared and source-checked with AI for Press-news Team. The source notes identify the material available during preparation. The article received no human editorial or clinical review. The source list supports the reported results and the calculations shown.
Source check: AI source check — primary research and stated source limits
Clinical review: Not applicable to this educational guide
Suggest a correction