Since manual detection of brain metastases (BMs) is time consuming, studies have been conducted to automate this process using deep learning. The purpose of this study was to conduct a systematic ...review and meta-analysis of the performance of deep learning models that use magnetic resonance imaging (MRI) to detect BMs in cancer patients. A systematic search of MEDLINE, EMBASE, and Web of Science was conducted until 30 September 2022. Inclusion criteria were: patients with BMs; deep learning using MRI images was applied to detect the BMs; sufficient data were present in terms of detective performance; original research articles. Exclusion criteria were: reviews, letters, guidelines, editorials, or errata; case reports or series with less than 20 patients; studies with overlapping cohorts; insufficient data in terms of detective performance; machine learning was used to detect BMs; articles not written in English. Quality Assessment of Diagnostic Accuracy Studies-2 and Checklist for Artificial Intelligence in Medical Imaging was used to assess the quality. Finally, 24 eligible studies were identified for the quantitative analysis. The pooled proportion of patient-wise and lesion-wise detectability was 89%. Articles should adhere to the checklists more strictly. Deep learning algorithms effectively detect BMs. Pooled analysis of false positive rates could not be estimated due to reporting differences.
•A very low rate of 3 ± 3 FPs per patient was maintained for all brain metastasis sizes.•Automated segmentation performance for brain metastases depends on metastasis size.•For metastases ≥6 mm, DSC ...was 87%, sensitivity was 99%, and PPV was 67%.•For metastases ≥3 mm and <6 mm, DSC was 64%, sensitivity was 87%, and PPV was 49%.•For metastases <3 mm, DSC was 17%, sensitivity was 25%, and PPV was 28%.
Brain metastases are manually contoured during stereotactic radiosurgery (SRS) treatment planning, which is time-consuming, potentially challenging, and laborious. The purpose of this study was to develop and investigate a 2-stage deep learning (DL) approach (MetNet) for brain metastasis segmentation in pre-treatment magnetic resonance imaging (MRI).
We retrospectively analyzed postcontrast 3D T1-weighted spoiled gradient echo MRIs from 934 patients who underwent SRS between August 2009 and August 2018. Neuroradiologists manually identified brain metastases in the MRIs. The treating radiation oncologist or physicist contoured the brain metastases. We constructed a 2-stage DL ensemble consisting of detection and segmentation models to segment the brain metastases on the MRIs. We evaluated the performance of MetNet by computing sensitivity, positive predictive value (PPV), and Dice similarity coefficient (DSC) with respect to metastasis size, as well as free-response receiver operating characteristics.
The 934 patients (mean ±standard deviation age 59 ± 13 years, 474 women) were randomly split into 80% training and 20% testing groups (748:186). For patients with metastases 1–52 mm (n = 766), 648 (85%) were detected and segmented with a mean segmentation DSC of 81% ± 15%. Patient-averaged sensitivity was 88% ± 19%, PPV was 58% ± 25%, and DSC was 85% ± 13% with 3 ± 3 false positives (FPs) per patient. When considering only metastases ≥6 mm, patient-averaged sensitivity was 99% ± 5%, PPV was 67% ± 28%, and DSC was 87% ± 13% with 1 ± 2 FPs per patient.
MetNet can segment brain metastases across a broad range of metastasis sizes with high sensitivity, low FPs, and high segmentation accuracy in postcontrast T1-weighted MRI, potentially aiding treatment planning for SRS.
In the present study, we have presented and validated a plastic scintillation detector (PSD) system designed for real-time multiprobe in vivo measurements.
The PSDs were built with a dose-sensitive ...volume of 0.4 mm(3). The PSDs were assembled into modular detector patches, each containing five closely packed PSDs. Continuous dose readings were performed every 150 ms, with a gap between consecutive readings of <0.3 ms. We first studied the effect of electron multiplication. We then assessed system performance in acrylic and anthropomorphic pelvic phantoms.
The PSDs were compatible with clinical rectal balloons and were easily inserted into the anthropomorphic phantom. With an electron multiplication average gain factor of 40, a twofold increase in the signal/noise ratio was observed, making near real-time dosimetry feasible. Under calibration conditions, the PSDs agreed with the ion chamber measurements to 0.08%. Precision, evaluated as a function of the total dose delivered, ranged from 2.3% at 2 cGy to 0.4% at 200 cGy.
Real-time PSD measurements are highly accurate and precise. These PSDs can be mounted onto rectal balloons, transforming these clinical devices into in vivo dose detectors without modifying current clinical practice. Real-time monitoring of the dose delivered near the rectum during prostate radiotherapy should help radiation oncologists protect this sensitive normal structure.
A strong dose-volume relationship exists between the amount of small bowel receiving low- to intermediate-doses of radiation and the rates of acute, severe gastrointestinal toxicity, principally ...diarrhea. There is considerable interest in the application of highly conformal treatment approaches, such as intensity-modulated radiation therapy (IMRT), to reduce dose to adjacent organs-at-risk in the treatment of carcinoma of the rectum. Therefore, we performed a comprehensive dosimetric evaluation of IMRT compared to 3-dimensional conformal radiation therapy (3DCRT) in standard, preoperative treatment for rectal cancer.
Using RTOG consensus anorectal contouring guidelines, treatment volumes were generated for ten patients treated preoperatively at our institution for rectal carcinoma, with IMRT plans compared to plans derived from classic anatomic landmarks, as well as 3DCRT plans treating the RTOG consensus volume. The patients were all T3, were node-negative (N = 1) or node-positive (N = 9), and were planned to a total dose of 45-Gy. Pairwise comparisons were made between IMRT and 3DCRT plans with respect to dose-volume histogram parameters.
IMRT plans had superior PTV coverage, dose homogeneity, and conformality in treatment of the gross disease and at-risk nodal volume, in comparison to 3DCRT. Additionally, in comparison to the 3DCRT plans, IMRT achieved a concomitant reduction in doses to the bowel (small bowel mean dose: 18.6-Gy IMRT versus 25.2-Gy 3DCRT; p = 0.005), bladder (V40Gy: 56.8% IMRT versus 75.4% 3DCRT; p = 0.005), pelvic bones (V40Gy: 47.0% IMRT versus 56.9% 3DCRT; p = 0.005), and femoral heads (V40Gy: 3.4% IMRT versus 9.1% 3DCRT; p = 0.005), with an improvement in absolute volumes of small bowel receiving dose levels known to induce clinically-relevant acute toxicity (small bowel V15Gy: 138-cc IMRT versus 157-cc 3DCRT; p = 0.005). We found that the IMRT treatment volumes were typically larger than that covered by classic bony landmark-derived fields, without incurring penalty with respect to adjacent organs-at-risk.
For rectal carcinoma, IMRT, compared to 3DCRT, yielded plans superior with respect to target coverage, homogeneity, and conformality, while lowering dose to adjacent organs-at-risk. This is achieved despite treating larger volumes, raising the possibility of a clinically-relevant improvement in the therapeutic ratio through the use of IMRT with a belly-board apparatus.
Compared with photon cranial radiation therapy (X-CRT), proton cranial radiation therapy (P-CRT) offers potential advantages in limiting radiation-induced sequalae in the treatment of pediatric brain ...tumors. This study aims to identify cognitive, functional magnetic resonance and positron emission tomography imaging markers and molecular differences between the radiation modalities.
Juvenile rats received a single faction of 10 Gy (relative biological effectiveness–weighted dose) delivered with 6 MV X-CRT or at the midspread out Bragg peak of a 100 MeV P-CRT beam. At 3, 6, and 12 months post-CRT, executive function was measured using 5-choice serial reaction time task. At ∼12 months post-CRT, animals were imaged with 18F-Flurodeoxy-glucose positron emission tomography imaging followed by functional ex vivo magnetic resonance imaging and stained for markers of neuroinflammation.
Irradiated animals had cognitive impairment with a higher number of omissions and lower incorrect and premature responses compared with sham (P ≤ .05). The accuracy of the animals’ X-CRT was less than that of sham (P ≤ .001). No significant difference in rates of cognitive change were found between the radiation modalities. At 12 months post-CRT, glucose metabolism was significantly higher than sham in X-CRT (P = .04) but not P-CRT. Using diffusion tensor imaging, P-CRT brains were found to have higher white matter volume and fiber lengths compared with sham (P < .03). Only X-CRT animals had higher apparent diffusion coefficient values compared with sham (P = .04). P-CRT animals had more connectomic changes compared with X-CRT. Correlative analysis identified several imaging features with cognitive performance. Furthermore, microgliosis (P < .05), astrogliosis (P < .01), and myelin thinning (P <.05) were observed in both radiation modalities, with X-CRT showing slightly more inflammation.
Both P-CRT and X-CRT lead to neurocognitive changes compared with sham. Although no significant difference was observed in neuroinflammation between the irradiated groups, differences were found in late-term glucose metabolism and brain connectome. Our results indicate that despite relative biological effectiveness weighting of the proton dose there are still differential effects which warrants further investigation.
Purpose
We investigated the feasibility of thoracic spine stereotactic body radiotherapy (SBRT) using the Elekta Unity magnetic resonance‐guided linear accelerator (MRL) in patients who received ...prior radiotherapy. We hypothesized that Monaco treatment plans can improve the gross tumor volume minimum dose (GTVmin) with spinal cord preservation and maintain consistent plan quality during daily adaptation.
Methods
Pinnacle clinical plans for 10 patients who underwent thoracic spine SBRT (after prior radiotherapy) were regenerated in the Monaco treatment planning system for the Elekta Unity MRL using 9 and 13 intensity‐modulated radiotherapy (IMRT) beams. Monaco adapt‐to‐position (ATP) and adapt‐to‐shape (ATS) workflow plans were generated using magnetic resonance imaging with a simulated daily positional setup deviation, and these adaptive plans were compared with Monaco reference plans. Plan quality measures included target coverage, Paddick conformity index, gradient index, homogeneity index, spinal cord D0.01cc, esophagus D0.01cc, lung V10, and skin D0.01cc.
Results
GTVmin values from the Monaco 9‐beam and 13‐beam plans were significantly higher than those from Pinnacle plans (p < 0.01) with similar spinal cord dose. Spinal cord D0.01cc, esophagus D0.01cc, and lung V10 did not statistically differ among the three plans. The electron‐return effect did not induce remarkable dose effects around the lungs or skin. While in the ATP workflow, a large increase in GTVmin was observed at the cost of a 10%–50% increase in spinal cord D0.01cc, in the ATS workflow, the spinal cord dose increase was maintained within 3% of the reference plan.
Conclusion
These findings show that MRL plans for thoracic spine SBRT are safe and feasible, allowing tumor dose escalation with spinal cord preservation and consistent daily plan adaptation using the ATS workflow. Careful plan review of hot spots and lung dose is necessary for safe MRL‐based treatment.
To determine whether there exists any significant difference in normal tissue toxicity between intensity modulated radiation therapy (IMRT) or proton therapy for the treatment of non-small cell lung ...cancer.
A total of 134 study patients (n=49 treated with proton therapy, n=85 with IMRT) treated in a randomized trial had a previously validated esophageal toxicity imaging biomarker, esophageal expansion, quantified during radiation therapy, as well as esophagitis grade (Common Terminology Criteria for Adverse Events version 3.0), on a weekly basis during treatment. Differences between the 2 modalities were statically analyzed using the imaging biomarker metric value (Kruskal-Wallis analysis of variance), as well as the incidence and severity of esophagitis grade (χ
and Fisher exact tests, respectively). The dose-response of the imaging biomarker was also compared between modalities using esophageal equivalent uniform dose, as well as delivered dose to an isotropic esophageal subvolume.
No statistically significant difference in the distribution of esophagitis grade, the incidence of grade ≥3 esophagitis (15 and 11 patients treated with IMRT and proton therapy, respectively), or the esophageal expansion imaging biomarker between cohorts (P>.05) was found. The distribution of imaging biomarker metric values had similar distributions between treatment arms, despite a slightly higher dose volume in the proton arm (P>.05). Imaging biomarker dose-response was similar between modalities for dose quantified as esophageal equivalent uniform dose and delivered esophageal subvolume dose. Regardless of treatment modality, there was high variability in imaging biomarker response, as well as esophagitis grade, for similar esophageal doses between patients.
There was no significant difference in esophageal toxicity from either proton- or photon-based radiation therapy as quantified by esophagitis grade or the esophageal expansion imaging biomarker.
Purpose
To develop an automated workflow for rectal cancer three‐dimensional conformal radiotherapy (3DCRT) treatment planning that combines deep learning (DL) aperture predictions and ...forward‐planning algorithms.
Methods
We designed an algorithm to automate the clinical workflow for 3DCRT planning with field aperture creations and field‐in‐field (FIF) planning. DL models (DeepLabV3+ architecture) were trained, validated, and tested on 555 patients to automatically generate aperture shapes for primary (posterior–anterior PA and opposed laterals) and boost fields. Network inputs were digitally reconstructed radiographs, gross tumor volume (GTV), and nodal GTV. A physician scored each aperture for 20 patients on a 5‐point scale (>3 is acceptable). A planning algorithm was then developed to create a homogeneous dose using a combination of wedges and subfields. The algorithm iteratively identifies a hotspot volume, creates a subfield, calculates dose, and optimizes beam weight all without user intervention. The algorithm was tested on 20 patients using clinical apertures with varying wedge angles and definitions of hotspots, and the resulting plans were scored by a physician. The end‐to‐end workflow was tested and scored by a physician on another 39 patients.
Results
The predicted apertures had Dice scores of 0.95, 0.94, and 0.90 for PA, laterals, and boost fields, respectively. Overall, 100%, 95%, and 87.5% of the PA, laterals, and boost apertures were scored as clinically acceptable, respectively. At least one auto‐plan was clinically acceptable for all patients. Wedged and non‐wedged plans were clinically acceptable for 85% and 50% of patients, respectively. The hotspot dose percentage was reduced from 121% (σ = 14%) to 109% (σ = 5%) of prescription dose for all plans. The integrated end‐to‐end workflow of automatically generated apertures and optimized FIF planning gave clinically acceptable plans for 38/39 (97%) of patients.
Conclusion
We have successfully automated the clinical workflow for generating radiotherapy plans for rectal cancer for our institution.
PDF
