Purpose

To explore the performance of deep learning to perform fully automated segmentation of rectal tumors on baseline MRI using a multicenter dataset (with variations in acquisition protocols and scan quality reflecting daily practice).

Material and methods

Baseline staging MRIs of 355 patients (from 6 centers) were analyzed. Overall scan quality was assessed using a 5-point Likert score (0=poor) to 4=excellent). Data were randomly split into train:validation:test cohorts (ratio of 5:1:4). An expert-radiologist manually delineated all rectal tumours to serve as training input. Using the T2-weighted, diffusion-weighted imaging (DWI)-b1000 and apparent diffusion coefficient (ADC)-maps from the training+validation cohorts, an attention-gated u-net was trained, with optimal hyperparameters (including learning rate, optimizer algorithm and batch-size) determined via grid search. A second expert-radiologist independently re-segmented all patients in the test cohort to calculate inter-reader agreement using the dice similarity coefficient (DSC). Agreement (DSC) between the trained network and the expert-segmentations was compared to the expert inter-reader agreement (serving as the standard of reference).

Results

Average DSC between expert-readers was 0.75 (±0.18). Average DSC for the network-generated segmentations was 0.64 (±0.22) and 0.59 (±0.22) compared to expert-reader 1 and 2, respectively. When excluding poor quality (score ≤ 2) scans, DSC between the network and expert-readers increased to 0.68 (±0.17) and 0.63 (±0.17).

Conclusion

Despite large variations in scan protocol/image quality, deep learning networks achieved a promising overall performance to automatically segment rectal tumours on MRI. Better results were achieved after exclusion of poor-quality scans, with agreement levels up to 0.63-0.68 (versus 0.75 between expert-radiologists).

Purpose

To evaluate whether MRI is accurate to identify poor responders after chemoradiotherapy who will need straight surgery and to evaluate whether results are reproducible amongst radiologists with different levels of expertise.

Material and methods

Seven independent readers with different expertise retrospectively evaluated the restaging MRIs (T2W+DWI) of 62 patients to categorize them as [1] poor responders—highly suspicious of tumour, [2] intermediate responders—tumour most likely, and [3] good—potential (near) complete responders. Reference standard was histopathology after surgery (or long-term follow-up in case of a watch-and-wait program).

Results

Fourteen patients were complete responders, 48 had residual tumour. Median percentage of patients categorized as “poor”, “intermediate” and “good” responders by the 7 readers was 21% (range 11-37%), 50% (range 23-58%) and 29% (range 23-42%). The vast majority of the poor responders had histopathologically confirmed residual tumour (of which 73% ypT3-4) with a low rate (0-5%) of “missed complete responders”. Of the 14 confirmed complete responders, a median percentage of 71% were categorized in the MR-good response and 29% in the MR-intermediate response group.

Conclusion

Radiologists of varying experience levels should be able to use MRI to identify the subgroup of ±20% of poor responding patients who will unavoidably require surgical resection after CRT. This may facilitate a more selective use of endoscopy, particularly in general settings or in centers with limited access to endoscopy.

Learning objectives

Learning objectives

Purpose

To evaluate the clinical impact of dedicated whole-body MRI (WB-MRI) in patients with stage IV colorectal cancer (CRC) on CT.

Material and methods

This was a retrospective study of 42 patients with CRC referred to our specialist oncologic center with suspicion of liver and/or peritoneal metastasis. All patients underwent an additional 3T WB-MRI at our institution, including pre/postcontrast T1, T2 and diffusion weighted imaging (DWI). Multidisciplinary team (MDT) meeting notes were evaluated and if the result of WB-MRI changed the original treatment plan (based on the CT result), it was regarded as a significant clinical impact.

Results

All metastases detected on CT images were also detected with WB-MRI. WB-MRI changed the treatment strategy in 13 patients (31%). WB-MRI diagnosed additional liver and peritoneal metastases in 3 (7%) and 1 (2%) patients, respectively. WB-MRI confirmed or ruled out liver and peritoneal metastases in indeterminate CT studies in 1 (2%) and 6 (14%) patients respectively. In one patient (2%), WB-MRI confirmed pulmonary metastasis in a lesion that was indeterminate with CT and in another patient (2%) WB-MRI detected an additional metastatic lesion in the anterior abdominal wall and potential renal malignancy.

Conclusion

Dedicated WB-MRI changed the treatment plan in almost one third (31%) of patients with stage IV CRC on CT. WB-MRI should be considered in this patient group to prevent undertreatment.

Learning objectives

Learning objectives

Purpose

To explore the performance of deep learning to perform fully automated segmentation of rectal tumors on baseline MRI using a multicenter dataset (with variations in acquisition protocols and scan quality reflecting daily practice).

Material and methods

Baseline staging MRIs of 355 patients (from 6 centers) were analyzed. Overall scan quality was assessed using a 5-point Likert score (0=poor) to 4=excellent). Data were randomly split into train:validation:test cohorts (ratio of 5:1:4). An expert-radiologist manually delineated all rectal tumours to serve as training input. Using the T2-weighted, diffusion-weighted imaging (DWI)-b1000 and apparent diffusion coefficient (ADC)-maps from the training+validation cohorts, an attention-gated u-net was trained, with optimal hyperparameters (including learning rate, optimizer algorithm and batch-size) determined via grid search. A second expert-radiologist independently re-segmented all patients in the test cohort to calculate inter-reader agreement using the dice similarity coefficient (DSC). Agreement (DSC) between the trained network and the expert-segmentations was compared to the expert inter-reader agreement (serving as the standard of reference).

Results

Average DSC between expert-readers was 0.75 (±0.18). Average DSC for the network-generated segmentations was 0.64 (±0.22) and 0.59 (±0.22) compared to expert-reader 1 and 2, respectively. When excluding poor quality (score ≤ 2) scans, DSC between the network and expert-readers increased to 0.68 (±0.17) and 0.63 (±0.17).

Conclusion

Despite large variations in scan protocol/image quality, deep learning networks achieved a promising overall performance to automatically segment rectal tumours on MRI. Better results were achieved after exclusion of poor-quality scans, with agreement levels up to 0.63-0.68 (versus 0.75 between expert-radiologists).

Purpose

The ESGAR-guidelines on MRI of rectal cancer advise the use of specific nodal staging criteria to determine the N-stage for primary rectal cancer staging, mainly aiming to avoid overstaging. These criteria were adapted from the Dutch national colorectal cancer guidelines that were introduced in 2014. The aim was to explore the clinical impact of the implementation of these Dutch guidelines on primary nodal staging outcomes in the Netherlands.

Material and methods

The primary staging MRIs of n=96 rectal cancer patients (from 3 Dutch centers) were analyzed: 48 patients from <2014 (pre-guideline) and 48 from >2014 (post-guideline). A dedicated reader determined the N-stage (N0/N1/N2) for each case, blinded to the original reports, using the criteria from the Dutch/ESGAR guidelines, where nodes are considered positive when ≥9mm, 5-8 mm with two morphologically suspicious criteria (round/irregular/heterogeneous), or <5 mm with 3 suspicious criteria. Results were compared to the N-stage in the original clinical reports, derived from the hospitals’ patient databases.

Results

Before 2014, the N-stage determined using the Dutch/ESGAR criteria was concordant with the original reports in 79% of the cases, the remaining 21% were downstaged when the guideline was applied. After 2014, scorings were concordant in 96% of the cases, 2% were upstaged, and 2% were downstaged. The difference in concordant/discrepant findings before and after 2014 was significant (P=0.014 Chi-square).

Conclusion

The results of this exploratory study suggest that the introduction of more strict nodal staging criteria has led to a significant reduction in overstaging of the nodal status in rectal cancer in the Netherlands.

Purpose

To evaluate whether MRI is accurate to identify poor responders after chemoradiotherapy who will need straight surgery and to evaluate whether results are reproducible amongst radiologists with different levels of expertise.

Material and methods

Seven independent readers with different expertise retrospectively evaluated the restaging MRIs (T2W+DWI) of 62 patients to categorize them as [1] poor responders—highly suspicious of tumour, [2] intermediate responders—tumour most likely, and [3] good—potential (near) complete responders. Reference standard was histopathology after surgery (or long-term follow-up in case of a watch-and-wait program).

Results

Fourteen patients were complete responders, 48 had residual tumour. Median percentage of patients categorized as “poor”, “intermediate” and “good” responders by the 7 readers was 21% (range 11-37%), 50% (range 23-58%) and 29% (range 23-42%). The vast majority of the poor responders had histopathologically confirmed residual tumour (of which 73% ypT3-4) with a low rate (0-5%) of “missed complete responders”. Of the 14 confirmed complete responders, a median percentage of 71% were categorized in the MR-good response and 29% in the MR-intermediate response group.

Conclusion

Radiologists of varying experience levels should be able to use MRI to identify the subgroup of ±20% of poor responding patients who will unavoidably require surgical resection after CRT. This may facilitate a more selective use of endoscopy, particularly in general settings or in centers with limited access to endoscopy.

Learning objectives