https://rauscha.github.io/Dialog-podcast A weekly peer-level digest of the most important new research on AI in medicine, focused on obstetrics and gynecology and on medical imaging, with Juno and Caspar. en Juno & Caspar false Juno & Caspar podcast@andrewrausch.com
Juno, Caspar, and any guest voices are AI-generated. Episode text and audio are generated with human-directed software.

Chapters:

1. 0:00 — Cold Open: What This Week's Headline Changes
2. 0:53 — Headline: Citation, Design, and Why the RCT Backbone Matters
3. 1:45 — Headline: Effect Sizes and Subgroup Interactions
4. 2:38 — Headline: Practice Impact and the Honest Caveat
5. 3:30 — Round 1: Maiter, Lung Cancer AI Head-to-Head
6. 4:23 — Round 2: Bahl, Breast Tomosynthesis Slab Reconstruction
7. 5:15 — Round 3: Zhang, Mamba Architecture pCR Prediction
8. 6:08 — Round 4: Liu, Radiomics for CMS4 Colorectal Subtyping
9. 7:01 — Close: Open Questions and Sign-Off

Follow-up links:

• Reperfusion Therapy in ESCAPE-MeVO Trial Participants: Imaging Characteristics and Clinical Outcomes - RCT-anchored imaging subgroup analysis in MeVO stroke; informs thrombectomy patient selection
• Independent Head-to-Head Comparison of Commercial Artificial Intelligence Devices for Lung Cancer Detection on Chest Radiographs - Independent benchmarking of commercial AI tools; directly informs deployment decisions
• Impact of AI-based Slab Reconstruction Technology on the Diagnostic Accuracy of Screening Digital Breast Tomosynthesis - Workflow optimization in breast imaging screening with maintained diagnostic performance
• Deep learning prediction of pathological complete response in breast cancer using Mamba architecture - Multicenter external validation of pCR prediction; relevant to neoadjuvant treatment planning
• Interpretable MRI-based Multiparametric Radiomics for Preoperative Prediction of CMS4 Colorectal Cancer - Preoperative molecular subtyping could inform surgical and systemic therapy planning

Sources:

• Ospel et al. - Radiology - 2026 - Reperfusion Therapy in ESCAPE-MeVO Trial Participants: Imaging Characteristics and Clinical Outcomes (https://doi.org/10.1148/radiol.251769)
• Maiter et al. - Radiology - 2026 - Independent Head-to-Head Comparison of Commercial Artificial Intelligence Devices for Lung Cancer Detection on Chest Radiographs (https://doi.org/10.1148/radiol.252205)
• Bahl et al. - Radiology - 2026 - Impact of AI-based Slab Reconstruction Technology on the Diagnostic Accuracy of Screening Digital Breast Tomosynthesis (https://doi.org/10.1148/radiol.252685)
• Zhang et al. - NPJ Digital Medicine - 2026 - Deep learning prediction of pathological complete response in breast cancer using Mamba architecture (https://doi.org/10.1038/s41746-026-02849-2)
• Liu et al. - Radiology - 2026 - Interpretable MRI-based Multiparametric Radiomics for Preoperative Prediction of CMS4 Colorectal Cancer (https://doi.org/10.1148/radiol.251719)

Original synthetic theme music generated locally for this episode.

Episode companion data]]>
Juno, Caspar, and any guest voices are AI-generated. Episode text and audio are generated with human-directed software.

Chapters:

1. 0:00 — Cold Open: What This Week's Headline Changes
2. 0:53 — Headline: Citation, Design, and Why the RCT Backbone Matters
3. 1:45 — Headline: Effect Sizes and Subgroup Interactions
4. 2:38 — Headline: Practice Impact and the Honest Caveat
5. 3:30 — Round 1: Maiter, Lung Cancer AI Head-to-Head
6. 4:23 — Round 2: Bahl, Breast Tomosynthesis Slab Reconstruction
7. 5:15 — Round 3: Zhang, Mamba Architecture pCR Prediction
8. 6:08 — Round 4: Liu, Radiomics for CMS4 Colorectal Subtyping
9. 7:01 — Close: Open Questions and Sign-Off

Follow-up links:

• Reperfusion Therapy in ESCAPE-MeVO Trial Participants: Imaging Characteristics and Clinical Outcomes - RCT-anchored imaging subgroup analysis in MeVO stroke; informs thrombectomy patient selection
• Independent Head-to-Head Comparison of Commercial Artificial Intelligence Devices for Lung Cancer Detection on Chest Radiographs - Independent benchmarking of commercial AI tools; directly informs deployment decisions
• Impact of AI-based Slab Reconstruction Technology on the Diagnostic Accuracy of Screening Digital Breast Tomosynthesis - Workflow optimization in breast imaging screening with maintained diagnostic performance
• Deep learning prediction of pathological complete response in breast cancer using Mamba architecture - Multicenter external validation of pCR prediction; relevant to neoadjuvant treatment planning
• Interpretable MRI-based Multiparametric Radiomics for Preoperative Prediction of CMS4 Colorectal Cancer - Preoperative molecular subtyping could inform surgical and systemic therapy planning

Sources:

• Ospel et al. - Radiology - 2026 - Reperfusion Therapy in ESCAPE-MeVO Trial Participants: Imaging Characteristics and Clinical Outcomes (https://doi.org/10.1148/radiol.251769)
• Maiter et al. - Radiology - 2026 - Independent Head-to-Head Comparison of Commercial Artificial Intelligence Devices for Lung Cancer Detection on Chest Radiographs (https://doi.org/10.1148/radiol.252205)
• Bahl et al. - Radiology - 2026 - Impact of AI-based Slab Reconstruction Technology on the Diagnostic Accuracy of Screening Digital Breast Tomosynthesis (https://doi.org/10.1148/radiol.252685)
• Zhang et al. - NPJ Digital Medicine - 2026 - Deep learning prediction of pathological complete response in breast cancer using Mamba architecture (https://doi.org/10.1038/s41746-026-02849-2)
• Liu et al. - Radiology - 2026 - Interpretable MRI-based Multiparametric Radiomics for Preoperative Prediction of CMS4 Colorectal Cancer (https://doi.org/10.1148/radiol.251719)

Original synthetic theme music generated locally for this episode.

Episode companion data]]> Wed, 03 Jun 2026 05:58:49 +0000 8fe782935698398b4308245f1ca4ea958c9c45234e4bb32c187b76d1fd7e7041 8:38 false
Juno, Caspar, and any guest voices are AI-generated. Episode text and audio are generated with human-directed software.

Chapters:

1. 0:00 — Cold Open: What Changed This Week
2. 0:02 — Headline: Citation and Design
3. 0:03 — Headline: Effect and Numbers
4. 0:04 — Headline: The Operator-Dependence Caveat (Affectionate Disagreement)
5. 0:06 — Headline: Practice Impact
6. 0:08 — Round 1: Asmara (Lung Nodule AI Benchmarks)
7. 0:09 — Round 2: Chen (Cardiac MRI, MVO, MACE Prediction)
8. 0:10 — Round 3: Quattrone (PSP Midbrain Index)
9. 0:12 — Round 4: Shin (LI-RADS v2024)
10. 0:14 — Close: What to Watch, What's Unsettled

Follow-up links:

• Prospective comparison of diagnostic accuracy of ultrasound, PET/CT and DW-MRI for preoperative assessment of pelvic lymph nodes in cervical cancer patients: results of the CANNES trial - First prospective multicenter trial demonstrating ultrasound non-inferiority to PET/CT and DW-MRI for pelvic node staging in cervical cancer — directly informs resource allocation and staging protocols
• Externally Tested AI Models for Malignancy Classification of Lung Nodules at Chest CT: A Systematic Review and Meta-Analysis - Provides pooled external validation benchmarks for AI lung nodule classifiers — essential for understanding real-world performance expectations
• AI-based Histologic Heterogeneity of Microvascular Obstruction at Cardiac MRI for Predicting MACEs: A Multicenter Study - Demonstrates AI radiomic features from automated MVO segmentation can outperform manual quantification for MACE prediction — early signal for prognostic AI in cardiac imaging
• Planimetric and Linear MRI Markers for Progressive Supranuclear Palsy Classification: A Large Multicohort International Study - Validates a simple dual-line midbrain index across 2,111 participants for PSP differentiation — methodologic model for imaging biomarker development
• Evaluating Accuracy of LI-RADS Nonradiation Treatment Response Algorithm v2024 and Ancillary Features at Hepatobiliary MRI versus CT - Confirms updated LI-RADS TRA v2024 with ancillary features improves HCC viability assessment — relevant for structured reporting algorithm evolution

Episode companion data]]>
Juno, Caspar, and any guest voices are AI-generated. Episode text and audio are generated with human-directed software.

Chapters:

1. 0:00 — Cold Open: What Changed This Week
2. 0:02 — Headline: Citation and Design
3. 0:03 — Headline: Effect and Numbers
4. 0:04 — Headline: The Operator-Dependence Caveat (Affectionate Disagreement)
5. 0:06 — Headline: Practice Impact
6. 0:08 — Round 1: Asmara (Lung Nodule AI Benchmarks)
7. 0:09 — Round 2: Chen (Cardiac MRI, MVO, MACE Prediction)
8. 0:10 — Round 3: Quattrone (PSP Midbrain Index)
9. 0:12 — Round 4: Shin (LI-RADS v2024)
10. 0:14 — Close: What to Watch, What's Unsettled

Follow-up links:

• Prospective comparison of diagnostic accuracy of ultrasound, PET/CT and DW-MRI for preoperative assessment of pelvic lymph nodes in cervical cancer patients: results of the CANNES trial - First prospective multicenter trial demonstrating ultrasound non-inferiority to PET/CT and DW-MRI for pelvic node staging in cervical cancer — directly informs resource allocation and staging protocols
• Externally Tested AI Models for Malignancy Classification of Lung Nodules at Chest CT: A Systematic Review and Meta-Analysis - Provides pooled external validation benchmarks for AI lung nodule classifiers — essential for understanding real-world performance expectations
• AI-based Histologic Heterogeneity of Microvascular Obstruction at Cardiac MRI for Predicting MACEs: A Multicenter Study - Demonstrates AI radiomic features from automated MVO segmentation can outperform manual quantification for MACE prediction — early signal for prognostic AI in cardiac imaging
• Planimetric and Linear MRI Markers for Progressive Supranuclear Palsy Classification: A Large Multicohort International Study - Validates a simple dual-line midbrain index across 2,111 participants for PSP differentiation — methodologic model for imaging biomarker development
• Evaluating Accuracy of LI-RADS Nonradiation Treatment Response Algorithm v2024 and Ancillary Features at Hepatobiliary MRI versus CT - Confirms updated LI-RADS TRA v2024 with ancillary features improves HCC viability assessment — relevant for structured reporting algorithm evolution

Sources:

• Frühauf et al. - Ultrasound Obstet Gynecol - 2026 - Prospective comparison of diagnostic accuracy of ultrasound, PET/CT and DW-MRI for preoperative assessment of pelvic lymph nodes in cervical cancer patients: results of the CANNES trial (https://doi.org/10.1002/uog.70241)
• Asmara et al. - Radiol Artif Intell - 2026 - Externally Tested AI Models for Malignancy Classification of Lung Nodules at Chest CT: A Systematic Review and Meta-Analysis (https://doi.org/10.1148/ryai.250331)
• Chen et al. - Radiology - 2026 - AI-based Histologic Heterogeneity of Microvascular Obstruction at Cardiac MRI for Predicting MACEs: A Multicenter Study (https://doi.org/10.1148/radiol.252250)
• Quattrone et al. - Radiology - 2026 - Planimetric and Linear MRI Markers for Progressive Supranuclear Palsy Classification: A Large Multicohort International Study (https://doi.org/10.1148/radiol.251394)
• Shin et al. - Radiology - 2026 - Evaluating Accuracy of LI-RADS Nonradiation Treatment Response Algorithm v2024 and Ancillary Features at Hepatobiliary MRI versus CT (https://doi.org/10.1148/radiol.253143)

Original synthetic theme music generated locally for this episode.

Episode companion data]]> Fri, 05 Jun 2026 10:08:12 +0000 dfed1996a4eeae827eef4d31938c32b266c350149e6ed187f790ea3392b34e69 0:37 false
Juno, Caspar, and any guest voices are AI-generated. Episode text and audio are generated with human-directed software.

Chapters:

1. 0:00 — Cold Open: The Annotation Problem, Named
2. 0:49 — Headline: Citation, Design, and the Choice That Earns the Evidence Grade
3. 1:37 — Headline: Numbers, Heterogeneity, and What "Six Cohorts" Does and Doesn't Prove
4. 2:26 — Headline: Practice Impact — Honest About "Nothing Yet"
5. 3:14 — Pivot and Round 1: Landry, Meningioma H&E Classification
6. 4:03 — Round 2: de Bie, Focal Laser Ablation Registry
7. 4:52 — Round 3: Kim, SKELEX Foundation Model
8. 5:40 — Round 4: Zhang, Paraspinal Muscle Fracture Prediction
9. 6:29 — Close: What to Watch, What's Unsettled, Sign-Off

Follow-up links:

• Enhancing foundation model transfer for prostate cancer detection with patch-level contrastive learning - Demonstrates that weakly supervised contrastive learning can maintain consistent prostate cancer detection across six heterogeneous MRI cohorts without pixel-level annotation
• Deep learning for H&E-based meningioma molecular classification and outcome prediction - Shows that routine histology slides can predict meningioma molecular subtypes and recurrence, potentially reducing reliance on genomic testing
• One-year follow-up after US-guided transperineal focal laser ablation of localized prostate cancer - Provides multicenter registry data on 12-month oncological outcomes for image-guided focal therapy, informing patient selection discussions
• A large-scale vision foundation model for musculoskeletal radiographs - SKELEX demonstrates that foundation models pretrained on over a million radiographs can outperform task-specific models across diverse MSK diagnostic tasks
• Fully automated system predicts osteoporotic vertebral fracture using lumbar MRI paraspinal muscle signatures - Automated paraspinal muscle analysis on MRI predicts vertebral fracture risk better than clinical models alone, suggesting opportunistic screening potential

Episode companion data]]>
Juno, Caspar, and any guest voices are AI-generated. Episode text and audio are generated with human-directed software.

Chapters:

1. 0:00 — Cold Open: The Annotation Problem, Named
2. 0:49 — Headline: Citation, Design, and the Choice That Earns the Evidence Grade
3. 1:37 — Headline: Numbers, Heterogeneity, and What "Six Cohorts" Does and Doesn't Prove
4. 2:26 — Headline: Practice Impact — Honest About "Nothing Yet"
5. 3:14 — Pivot and Round 1: Landry, Meningioma H&E Classification
6. 4:03 — Round 2: de Bie, Focal Laser Ablation Registry
7. 4:52 — Round 3: Kim, SKELEX Foundation Model
8. 5:40 — Round 4: Zhang, Paraspinal Muscle Fracture Prediction
9. 6:29 — Close: What to Watch, What's Unsettled, Sign-Off

Follow-up links:

• Enhancing foundation model transfer for prostate cancer detection with patch-level contrastive learning - Demonstrates that weakly supervised contrastive learning can maintain consistent prostate cancer detection across six heterogeneous MRI cohorts without pixel-level annotation
• Deep learning for H&E-based meningioma molecular classification and outcome prediction - Shows that routine histology slides can predict meningioma molecular subtypes and recurrence, potentially reducing reliance on genomic testing
• One-year follow-up after US-guided transperineal focal laser ablation of localized prostate cancer - Provides multicenter registry data on 12-month oncological outcomes for image-guided focal therapy, informing patient selection discussions
• A large-scale vision foundation model for musculoskeletal radiographs - SKELEX demonstrates that foundation models pretrained on over a million radiographs can outperform task-specific models across diverse MSK diagnostic tasks
• Fully automated system predicts osteoporotic vertebral fracture using lumbar MRI paraspinal muscle signatures - Automated paraspinal muscle analysis on MRI predicts vertebral fracture risk better than clinical models alone, suggesting opportunistic screening potential

Sources:

• Lee et al. - NPJ Digital Medicine - 2026 - Enhancing foundation model transfer for prostate cancer detection with patch-level contrastive learning (https://doi.org/10.1038/s41746-026-02831-y)
• Landry et al. - Lancet Digital Health - 2026 - Deep learning for H&E-based meningioma molecular classification and outcome prediction (https://doi.org/10.1016/j.landig.2026.100986)
• de Bie - Radiology - 2026 - One-year follow-up after US-guided transperineal focal laser ablation of localized prostate cancer (https://doi.org/10.1148/radiol.251658)
• Kim et al. - NPJ Digital Medicine - 2026 - A large-scale vision foundation model for musculoskeletal radiographs (https://doi.org/10.1038/s41746-026-02826-9)
• Zhang et al. - NPJ Digital Medicine - 2026 - Fully automated system predicts osteoporotic vertebral fracture using lumbar MRI paraspinal muscle signatures (https://doi.org/10.1038/s41746-026-02855-4)

Original synthetic theme music generated locally for this episode.

Episode companion data]]> Sat, 06 Jun 2026 10:09:17 +0000 2a76a95319984e9a61387d02b99469920e8db3e7ab94b57e01dd8aeae1b7f43d 8:02 false
Juno, Caspar, and any guest voices are AI-generated. Episode text and audio are generated with human-directed software.

Chapters:

1. 0:00 — Cold Open: The Benchmark Shift
2. 0:57 — Headline: Clinical Question and Study Design
3. 1:55 — Headline: Effect Size and What Parity Means
4. 2:52 — Headline: Practice Impact and Honest Limits
5. 3:50 — Round 1: HCC Contrast Agent Equivalence (Heo)
6. 4:47 — Round 2: Pancreatic Cancer Detection on CT (Yamaguchi)
7. 5:45 — Round 3: Air Pollution and Coronary Risk, Sex-Specific (Castillo)
8. 6:42 — Round 4: Net Water Uptake for Thrombectomy Selection (Lakhani)
9. 7:40 — Close: What to Watch, What's Still Unsettled

Follow-up links:

• Integrating ultrasound-CT-MR for preoperative multi-task prediction in ovarian cancer: achieving diagnostic parity with multidisciplinary team consensus - First large external validation showing multimodal AI matches MDT consensus across five preoperative ovarian cancer classification tasks — a credible benchmark for clinical translation in gyn-onc.
• Diagnostic Accuracy of Extracellular versus Hepatobiliary Contrast-enhanced MRI in LI-RADS Nonradiation Treatment Response Algorithm Version 2024 - Demonstrates equivalent accuracy between ECA-MRI and HBA-MRI for post-treatment HCC assessment, informing protocol flexibility in surveillance imaging.
• Deep Learning Detection of Direct and Indirect Imaging Findings Associated with Pancreatic Cancer at Contrast-enhanced and Noncontrast CT - Multiinstitutional DL model detects pancreatic cancer on both NCCT and CECT with physician-level performance, opening doors for opportunistic screening.
• Sex-Specific Associations between Long-term Air Pollution Exposure and Coronary Atherosclerosis at Cardiac CT - Large cohort data linking PM2.5 and NO2 to coronary plaque burden with sex-specific patterns — relevant for cardiovascular risk counseling.
• Ischemic Lesion Net Water Uptake Outperforms ASPECTS and CBF Less Than 30% in Predicting Futile Recanalization after Mechanical Thrombectomy - Net water uptake emerges as a superior imaging biomarker for thrombectomy patient selection, potentially reducing futile interventions.

Episode companion data]]>
Juno, Caspar, and any guest voices are AI-generated. Episode text and audio are generated with human-directed software.

Chapters:

1. 0:00 — Cold Open: The Benchmark Shift
2. 0:57 — Headline: Clinical Question and Study Design
3. 1:55 — Headline: Effect Size and What Parity Means
4. 2:52 — Headline: Practice Impact and Honest Limits
5. 3:50 — Round 1: HCC Contrast Agent Equivalence (Heo)
6. 4:47 — Round 2: Pancreatic Cancer Detection on CT (Yamaguchi)
7. 5:45 — Round 3: Air Pollution and Coronary Risk, Sex-Specific (Castillo)
8. 6:42 — Round 4: Net Water Uptake for Thrombectomy Selection (Lakhani)
9. 7:40 — Close: What to Watch, What's Still Unsettled

Follow-up links:

• Integrating ultrasound-CT-MR for preoperative multi-task prediction in ovarian cancer: achieving diagnostic parity with multidisciplinary team consensus - First large external validation showing multimodal AI matches MDT consensus across five preoperative ovarian cancer classification tasks — a credible benchmark for clinical translation in gyn-onc.
• Diagnostic Accuracy of Extracellular versus Hepatobiliary Contrast-enhanced MRI in LI-RADS Nonradiation Treatment Response Algorithm Version 2024 - Demonstrates equivalent accuracy between ECA-MRI and HBA-MRI for post-treatment HCC assessment, informing protocol flexibility in surveillance imaging.
• Deep Learning Detection of Direct and Indirect Imaging Findings Associated with Pancreatic Cancer at Contrast-enhanced and Noncontrast CT - Multiinstitutional DL model detects pancreatic cancer on both NCCT and CECT with physician-level performance, opening doors for opportunistic screening.
• Sex-Specific Associations between Long-term Air Pollution Exposure and Coronary Atherosclerosis at Cardiac CT - Large cohort data linking PM2.5 and NO2 to coronary plaque burden with sex-specific patterns — relevant for cardiovascular risk counseling.
• Ischemic Lesion Net Water Uptake Outperforms ASPECTS and CBF Less Than 30% in Predicting Futile Recanalization after Mechanical Thrombectomy - Net water uptake emerges as a superior imaging biomarker for thrombectomy patient selection, potentially reducing futile interventions.

Sources:

• Yu et al. - NPJ Digital Medicine - 2026 - Integrating ultrasound-CT-MR for preoperative multi-task prediction in ovarian cancer: achieving diagnostic parity with multidisciplinary team consensus (https://doi.org/10.1038/s41746-026-02875-0)
• Heo et al. - Radiology - 2026 - Diagnostic Accuracy of Extracellular versus Hepatobiliary Contrast-enhanced MRI in LI-RADS Nonradiation Treatment Response Algorithm Version 2024 (https://doi.org/10.1148/radiol.253854)
• Yamaguchi et al. - Radiology - 2026 - Deep Learning Detection of Direct and Indirect Imaging Findings Associated with Pancreatic Cancer at Contrast-enhanced and Noncontrast CT (https://doi.org/10.1148/radiol.253122)
• Castillo et al. - Radiology - 2026 - Sex-Specific Associations between Long-term Air Pollution Exposure and Coronary Atherosclerosis at Cardiac CT (https://doi.org/10.1148/radiol.252086)
• Lakhani et al. - Radiology - 2026 - Ischemic Lesion Net Water Uptake Outperforms ASPECTS and CBF Less Than 30% in Predicting Futile Recanalization after Mechanical Thrombectomy (https://doi.org/10.1148/radiol.252487)

Original synthetic theme music generated locally for this episode.

Episode companion data]]> Fri, 19 Jun 2026 10:09:18 +0000 35980bdbc3d532aeb49fa294ec596ba027605a07fb534af9a4467ad038098644 9:22 false
Juno, Caspar, and any guest voices are AI-generated. Episode text and audio are generated with human-directed software.

Chapters:

1. 0:00 — Cold Open: The Annotation Problem
2. 0:51 — Headline: Clinical Question and Design
3. 1:43 — Headline: Effect, Caveat, and Practice Impact
4. 2:34 — Pivot
5. 3:25 — Round 1: Lehman / Longitudinal AI Mammography Risk
6. 4:16 — Round 2: Peeters / LUNA25 Lung Nodule Challenge
7. 5:08 — Round 3: Ma / MR Lymphangiography Venous Suppression
8. 5:59 — Round 4: Shu / Shortcut Learning in Chest Radiograph AI
9. 6:50 — Close: What's Unsettled

Follow-up links:

• Reconstruction from multi-planar MRI with foundation models for uterine fibroid analysis - Demonstrates annotation-free 3D fibroid segmentation with cross-center generalization, potentially reducing manual contouring burden in surgical planning workflows.
• Longitudinal Analysis of Changes in Deep Learning Image-based Breast Cancer Risk Scores over Time - Shows AI mammography risk scores rise before cancer diagnosis in a large cohort, supporting dynamic rather than static risk stratification.
• Benchmarking of AI and Radiologists for Indeterminate Lung Nodule Malignancy Risk Estimation on Screening CT: The LUNA25 Challenge - Establishes that top AI systems match or exceed radiologist performance for indeterminate lung nodule risk on a standardized external benchmark.
• MR Lymphangiography for Diagnosis of Lower Extremity Lymphedema: Suppressing Venous Signal Interference Using Deep Learning - Improves diagnostic clarity in MR lymphangiography for lymphedema, relevant to post-surgical and post-radiation imaging in oncology.
• Impact of Exposure Parameters on Deep Learning Models in Chest Radiography and Implications for Deployment - Reveals that DL models exploit exposure parameters as shortcuts, a deployment-critical bias finding for any chest radiograph AI.

Episode companion data]]>
Juno, Caspar, and any guest voices are AI-generated. Episode text and audio are generated with human-directed software.

Chapters:

1. 0:00 — Cold Open: The Annotation Problem
2. 0:51 — Headline: Clinical Question and Design
3. 1:43 — Headline: Effect, Caveat, and Practice Impact
4. 2:34 — Pivot
5. 3:25 — Round 1: Lehman / Longitudinal AI Mammography Risk
6. 4:16 — Round 2: Peeters / LUNA25 Lung Nodule Challenge
7. 5:08 — Round 3: Ma / MR Lymphangiography Venous Suppression
8. 5:59 — Round 4: Shu / Shortcut Learning in Chest Radiograph AI
9. 6:50 — Close: What's Unsettled

Follow-up links:

• Reconstruction from multi-planar MRI with foundation models for uterine fibroid analysis - Demonstrates annotation-free 3D fibroid segmentation with cross-center generalization, potentially reducing manual contouring burden in surgical planning workflows.
• Longitudinal Analysis of Changes in Deep Learning Image-based Breast Cancer Risk Scores over Time - Shows AI mammography risk scores rise before cancer diagnosis in a large cohort, supporting dynamic rather than static risk stratification.
• Benchmarking of AI and Radiologists for Indeterminate Lung Nodule Malignancy Risk Estimation on Screening CT: The LUNA25 Challenge - Establishes that top AI systems match or exceed radiologist performance for indeterminate lung nodule risk on a standardized external benchmark.
• MR Lymphangiography for Diagnosis of Lower Extremity Lymphedema: Suppressing Venous Signal Interference Using Deep Learning - Improves diagnostic clarity in MR lymphangiography for lymphedema, relevant to post-surgical and post-radiation imaging in oncology.
• Impact of Exposure Parameters on Deep Learning Models in Chest Radiography and Implications for Deployment - Reveals that DL models exploit exposure parameters as shortcuts, a deployment-critical bias finding for any chest radiograph AI.

Sources:

• Wang et al. - NPJ Digital Medicine - 2026 - Reconstruction from multi-planar MRI with foundation models for uterine fibroid analysis (https://doi.org/10.1038/s41746-026-02780-6)
• Lehman et al. - Radiology - 2026 - Longitudinal Analysis of Changes in Deep Learning Image-based Breast Cancer Risk Scores over Time (https://doi.org/10.1148/radiol.253023)
• Peeters et al. - Radiology Artificial Intelligence - 2026 - Benchmarking of AI and Radiologists for Indeterminate Lung Nodule Malignancy Risk Estimation on Screening CT: The LUNA25 Challenge (https://doi.org/10.1148/ryai.260179)
• Ma et al. - Radiology - 2026 - MR Lymphangiography for Diagnosis of Lower Extremity Lymphedema: Suppressing Venous Signal Interference Using Deep Learning (https://doi.org/10.1148/radiol.251225)
• Shu et al. - Radiology Artificial Intelligence - 2026 - Impact of Exposure Parameters on Deep Learning Models in Chest Radiography and Implications for Deployment (https://doi.org/10.1148/ryai.250731)

Original synthetic theme music generated locally for this episode.

Episode companion data]]> Fri, 26 Jun 2026 10:18:52 +0000 83b40ce412506fff04254e71682d1fbd69f5a8af3aae4f509cec7302e97aa614 8:26 false
Juno, Caspar, and any guest voices are AI-generated. Episode text and audio are generated with human-directed software.

Chapters:

1. 0:00 — Cold Open: The Structural Problem
2. 0:51 — Headline: Design and Population
3. 1:41 — Headline: The Finding and the Definitional Caveat
4. 2:32 — Headline: Practice Implication
5. 3:22 — Pivot and Round 1: Kitaguchi (AI Surgical Overlay)
6. 4:13 — Round 2: Abiodun-Ojo (PSMA-11 vs Fluciclovine)
7. 5:03 — Round 3: Ranieri Guimaraes (SCOT-HEART Body Composition)
8. 5:54 — Round 4: Shin (H&E Neoadjuvant Response Prediction)
9. 6:44 — Close: What to Watch, What Is Unsettled

Follow-up links:

• Pelvic MRI in Endometrial Cancer Staging: A Retrospective Evaluation of the Impact of FIGO 2023 - Demonstrates that FIGO 2023 molecular criteria reduce MRI-pathology concordance, forcing a rethink of preoperative imaging workflows in endometrial cancer.
• Enhancing anatomical recognition by surgeons during pelvic lymph node dissection using artificial intelligence - Shows AI overlay can meaningfully improve intraoperative anatomical identification, with implications for reducing surgical complications in pelvic oncology.
• 18F-Fluciclovine or 68Ga-PSMA-11 PET/CT-guided Salvage Radiotherapy Changes in Postprostatectomy Biochemical Recurrence: Secondary Analysis of the EMPIRE-2 Trial - RCT-derived evidence that PSMA-11 PET/CT changes salvage RT planning more than fluciclovine, informing tracer selection in recurrent prostate cancer.
• Machine Learning Multiorgan Analysis of Coronary CT Angiography Body Composition, Myocardial Infarction, and Mortality in the SCOT-HEART Trial - Demonstrates that ML-extracted body composition from routine coronary CTA adds independent prognostic value for hard cardiovascular endpoints over 10 years.
• Predicting response to neoadjuvant therapy using artificial intelligence on digitized histopathology slides: a systematic review - Synthesizes the state of AI for neoadjuvant response prediction, highlighting promising accuracy but critical gaps in external validation and standardization.

Episode companion data]]>
Juno, Caspar, and any guest voices are AI-generated. Episode text and audio are generated with human-directed software.

Chapters:

1. 0:00 — Cold Open: The Structural Problem
2. 0:51 — Headline: Design and Population
3. 1:41 — Headline: The Finding and the Definitional Caveat
4. 2:32 — Headline: Practice Implication
5. 3:22 — Pivot and Round 1: Kitaguchi (AI Surgical Overlay)
6. 4:13 — Round 2: Abiodun-Ojo (PSMA-11 vs Fluciclovine)
7. 5:03 — Round 3: Ranieri Guimaraes (SCOT-HEART Body Composition)
8. 5:54 — Round 4: Shin (H&E Neoadjuvant Response Prediction)
9. 6:44 — Close: What to Watch, What Is Unsettled

Follow-up links:

• Pelvic MRI in Endometrial Cancer Staging: A Retrospective Evaluation of the Impact of FIGO 2023 - Demonstrates that FIGO 2023 molecular criteria reduce MRI-pathology concordance, forcing a rethink of preoperative imaging workflows in endometrial cancer.
• Enhancing anatomical recognition by surgeons during pelvic lymph node dissection using artificial intelligence - Shows AI overlay can meaningfully improve intraoperative anatomical identification, with implications for reducing surgical complications in pelvic oncology.
• 18F-Fluciclovine or 68Ga-PSMA-11 PET/CT-guided Salvage Radiotherapy Changes in Postprostatectomy Biochemical Recurrence: Secondary Analysis of the EMPIRE-2 Trial - RCT-derived evidence that PSMA-11 PET/CT changes salvage RT planning more than fluciclovine, informing tracer selection in recurrent prostate cancer.
• Machine Learning Multiorgan Analysis of Coronary CT Angiography Body Composition, Myocardial Infarction, and Mortality in the SCOT-HEART Trial - Demonstrates that ML-extracted body composition from routine coronary CTA adds independent prognostic value for hard cardiovascular endpoints over 10 years.
• Predicting response to neoadjuvant therapy using artificial intelligence on digitized histopathology slides: a systematic review - Synthesizes the state of AI for neoadjuvant response prediction, highlighting promising accuracy but critical gaps in external validation and standardization.

Sources:

• Avesani et al. - Radiology - 2026 - Pelvic MRI in Endometrial Cancer Staging: A Retrospective Evaluation of the Impact of FIGO 2023 (https://doi.org/10.1148/radiol.252037)
• Kitaguchi et al. - NPJ Digit Med - 2026 - Enhancing anatomical recognition by surgeons during pelvic lymph node dissection using artificial intelligence (https://doi.org/10.1038/s41746-026-02936-4)
• Abiodun-Ojo et al. - Radiology - 2026 - 18F-Fluciclovine or 68Ga-PSMA-11 PET/CT-guided Salvage Radiotherapy Changes in Postprostatectomy Biochemical Recurrence: Secondary Analysis of the EMPIRE-2 Trial (https://doi.org/10.1148/radiol.251874)
• Ranieri Guimaraes - Radiology - 2026 - Machine Learning Multiorgan Analysis of Coronary CT Angiography Body Composition, Myocardial Infarction, and Mortality in the SCOT-HEART Trial (https://doi.org/10.1148/radiol.251821)
• Shin et al. - NPJ Digit Med - 2026 - Predicting response to neoadjuvant therapy using artificial intelligence on digitized histopathology slides: a systematic review (https://doi.org/10.1038/s41746-026-02947-1)

Original synthetic theme music generated locally for this episode.

Episode companion data]]> Sat, 04 Jul 2026 10:17:22 +0000 2892d6dbf03d5895cda8e46a5c22799da49a37e8dec69b22dd390bc82994b67d 8:20 false
Juno, Caspar, and any guest voices are AI-generated. Episode text and audio are generated with human-directed software.

Chapters:

1. 0:00 — Cold Open: The Selection Problem
2. 0:57 — Headline: Citation and Design
3. 1:55 — Headline: The Finding and the Caveat
4. 2:52 — Headline: Practice Implication
5. 3:49 — Round 1: Cardiac MRI Entropy in Myocarditis (Wang)
6. 4:47 — Round 2: Noncontrast Abbreviated MRI for HCC Post-TACE (Jeon)
7. 5:44 — Round 3: Retinal Fundus AI for AF Prediction (Xu, NPJ)
8. 6:42 — Round 4: RECIBM Bone Metastasis Response Criteria (Xu, Radiology)
9. 7:39 — Close: What to Watch, What's Unsettled

Follow-up links:

• Hyperdense Capsule Sign at Noncontrast CT as an Indication for Middle Meningeal Artery Embolization for Nonacute Subdural Hematomas: A MAGIC-MT Trial Post Hoc Analysis - Identifies an imaging biomarker that may stratify which SDH patients benefit from MMA embolization, potentially rationalizing procedural selection.
• Entropy for Prediction of MACEs in Myocarditis: A Cardiac MRI-based Biomarker of Myocardial Tissue Heterogeneity - Adds a texture-based prognostic metric to cardiac MRI in myocarditis, independent of LGE and EF.
• Noncontrast Abbreviated MRI for Post-TACE Treatment Response Monitoring of Hepatocellular Carcinoma Based on Ancillary Features from LI-RADS - Explores a contrast-sparing surveillance option for HCC patients post-TACE, relevant for renal-impaired or high-frequency imaging populations.
• Prediction of incident atrial fibrillation from retinal fundus images using a multimodal foundation model - Demonstrates that retinal imaging encodes AF risk, validated across multiethnic cohorts, with potential for opportunistic screening.
• Response Evaluation Criteria in Bone Metastases: Performance and Association of Response Classifications with Survival Outcomes - Proposes new bone metastasis response criteria with stronger survival association than existing frameworks, addressing a gap in oncologic imaging.

Episode companion data]]>
Juno, Caspar, and any guest voices are AI-generated. Episode text and audio are generated with human-directed software.

Chapters:

1. 0:00 — Cold Open: The Selection Problem
2. 0:57 — Headline: Citation and Design
3. 1:55 — Headline: The Finding and the Caveat
4. 2:52 — Headline: Practice Implication
5. 3:49 — Round 1: Cardiac MRI Entropy in Myocarditis (Wang)
6. 4:47 — Round 2: Noncontrast Abbreviated MRI for HCC Post-TACE (Jeon)
7. 5:44 — Round 3: Retinal Fundus AI for AF Prediction (Xu, NPJ)
8. 6:42 — Round 4: RECIBM Bone Metastasis Response Criteria (Xu, Radiology)
9. 7:39 — Close: What to Watch, What's Unsettled

Follow-up links:

• Hyperdense Capsule Sign at Noncontrast CT as an Indication for Middle Meningeal Artery Embolization for Nonacute Subdural Hematomas: A MAGIC-MT Trial Post Hoc Analysis - Identifies an imaging biomarker that may stratify which SDH patients benefit from MMA embolization, potentially rationalizing procedural selection.
• Entropy for Prediction of MACEs in Myocarditis: A Cardiac MRI-based Biomarker of Myocardial Tissue Heterogeneity - Adds a texture-based prognostic metric to cardiac MRI in myocarditis, independent of LGE and EF.
• Noncontrast Abbreviated MRI for Post-TACE Treatment Response Monitoring of Hepatocellular Carcinoma Based on Ancillary Features from LI-RADS - Explores a contrast-sparing surveillance option for HCC patients post-TACE, relevant for renal-impaired or high-frequency imaging populations.
• Prediction of incident atrial fibrillation from retinal fundus images using a multimodal foundation model - Demonstrates that retinal imaging encodes AF risk, validated across multiethnic cohorts, with potential for opportunistic screening.
• Response Evaluation Criteria in Bone Metastases: Performance and Association of Response Classifications with Survival Outcomes - Proposes new bone metastasis response criteria with stronger survival association than existing frameworks, addressing a gap in oncologic imaging.

Sources:

• Weng et al. - Radiology - 2026 - Hyperdense Capsule Sign at Noncontrast CT as an Indication for Middle Meningeal Artery Embolization for Nonacute Subdural Hematomas: A MAGIC-MT Trial Post Hoc Analysis (https://doi.org/10.1148/radiol.251512)
• Wang et al. - Radiology - 2026 - Entropy for Prediction of MACEs in Myocarditis: A Cardiac MRI-based Biomarker of Myocardial Tissue Heterogeneity (https://doi.org/10.1148/radiol.251932)
• Jeon et al. - Radiology - 2026 - Noncontrast Abbreviated MRI for Post-TACE Treatment Response Monitoring of Hepatocellular Carcinoma Based on Ancillary Features from LI-RADS (https://doi.org/10.1148/radiol.252803)
• Xu et al. - NPJ Digital Medicine - 2026 - Prediction of incident atrial fibrillation from retinal fundus images using a multimodal foundation model (https://doi.org/10.1038/s41746-026-02969-9)
• Xu et al. - Radiology - 2026 - Response Evaluation Criteria in Bone Metastases: Performance and Association of Response Classifications with Survival Outcomes (https://doi.org/10.1148/radiol.260907)

Original synthetic theme music generated locally for this episode.

Episode companion data]]> Fri, 10 Jul 2026 10:20:51 +0000 efc43253d2bf25b649f49f13eefad60f6a7cb13950ac50caf289f619091b0e02 9:21 false
Juno, Caspar, and any guest voices are AI-generated. Episode text and audio are generated with human-directed software.

Chapters:

1. 0:00 — Cold Open: What This Week Changes
2. 0:59 — Headline: Clinical Question and Design Choice
3. 1:58 — Headline: Effect Size and Its Conditions
4. 2:57 — Headline: Practice Impact and Open Thread
5. 3:56 — Pivot and Round 1: Taib Mammography / Automation Bias
6. 4:55 — Round 2: Lin CXR Foundation Model
7. 5:54 — Round 3: Kim PROVISION-AF / AI-ECG in Atrial Fibrillation
8. 6:53 — Round 4: Chen ccRCC / H&E-Based Immunotherapy Prediction
9. 7:52 — Close: What to Watch, What's Unsettled

Follow-up links:

• Multi-omics fusion with machine learning enables robust prediction of treatment response in ovarian cancer for precision population health. - First multi-omics fusion model for ovarian cancer chemotherapy response with AUC 0.939; identifies novel chemoresistance drivers; needs prospective validation.
• Automation Bias in Action: Eye Tracking of Humans Reading Screening Mammograms with and without AI Prompts. - Eye-tracking evidence that incorrect AI prompts cause readers to miss cancers and over-call false positives; directly relevant to breast imaging workflow design.
• Transparent chest radiograph foundation model enables explainable human disease profiling. - Foundation model predicts over 1,000 phenotypes from CXR across three cohorts; demonstrates systemic disease profiling potential with explainability.
• Impact of an AI algorithm for multi-day prediction of incident atrial fibrillation on clinical decision-making: PROVISION-AF study. - Multinational simulation shows AI-ECG improves physician AF prediction from AUROC 0.57 to 0.65; modest but real decision-support benefit.
• Deep learning-based CD8+ T cell model for predicting prognosis and targeted immunotherapy benefit in ccRCC. - H&E-based deep learning infers CD8+ inflammation and predicts immunotherapy benefit in renal cell carcinoma; AUC ~0.78 across two multicenter cohorts.

Episode companion data]]>
Juno, Caspar, and any guest voices are AI-generated. Episode text and audio are generated with human-directed software.

Chapters:

1. 0:00 — Cold Open: What This Week Changes
2. 0:59 — Headline: Clinical Question and Design Choice
3. 1:58 — Headline: Effect Size and Its Conditions
4. 2:57 — Headline: Practice Impact and Open Thread
5. 3:56 — Pivot and Round 1: Taib Mammography / Automation Bias
6. 4:55 — Round 2: Lin CXR Foundation Model
7. 5:54 — Round 3: Kim PROVISION-AF / AI-ECG in Atrial Fibrillation
8. 6:53 — Round 4: Chen ccRCC / H&E-Based Immunotherapy Prediction
9. 7:52 — Close: What to Watch, What's Unsettled

Follow-up links:

• Multi-omics fusion with machine learning enables robust prediction of treatment response in ovarian cancer for precision population health. - First multi-omics fusion model for ovarian cancer chemotherapy response with AUC 0.939; identifies novel chemoresistance drivers; needs prospective validation.
• Automation Bias in Action: Eye Tracking of Humans Reading Screening Mammograms with and without AI Prompts. - Eye-tracking evidence that incorrect AI prompts cause readers to miss cancers and over-call false positives; directly relevant to breast imaging workflow design.
• Transparent chest radiograph foundation model enables explainable human disease profiling. - Foundation model predicts over 1,000 phenotypes from CXR across three cohorts; demonstrates systemic disease profiling potential with explainability.
• Impact of an AI algorithm for multi-day prediction of incident atrial fibrillation on clinical decision-making: PROVISION-AF study. - Multinational simulation shows AI-ECG improves physician AF prediction from AUROC 0.57 to 0.65; modest but real decision-support benefit.
• Deep learning-based CD8+ T cell model for predicting prognosis and targeted immunotherapy benefit in ccRCC. - H&E-based deep learning infers CD8+ inflammation and predicts immunotherapy benefit in renal cell carcinoma; AUC ~0.78 across two multicenter cohorts.

Sources:

• Chen et al. - NPJ Digit Med - 2026 - Multi-omics fusion with machine learning enables robust prediction of treatment response in ovarian cancer for precision population health. (https://doi.org/10.1038/s41746-026-02991-x)
• Taib et al. - Radiology - 2026 - Automation Bias in Action: Eye Tracking of Humans Reading Screening Mammograms with and without AI Prompts. (https://doi.org/10.1148/radiol.252590)
• Lin et al. - NPJ Digit Med - 2026 - Transparent chest radiograph foundation model enables explainable human disease profiling. (https://doi.org/10.1038/s41746-026-02990-y)
• Kim et al. - NPJ Digit Med - 2026 - Impact of an AI algorithm for multi-day prediction of incident atrial fibrillation on clinical decision-making: PROVISION-AF study. (https://doi.org/10.1038/s41746-026-02950-6)
• Chen et al. - NPJ Digit Med - 2026 - Deep learning-based CD8+ T cell model for predicting prognosis and targeted immunotherapy benefit in ccRCC. (https://doi.org/10.1038/s41746-026-02955-1)

Original synthetic theme music generated locally for this episode.

Episode companion data]]> Fri, 17 Jul 2026 10:18:34 +0000 9bbac91295c9af7567da3eb983ec31d9f2525ec8683e6c3471770ea5e23be684 9:36 false