Datasets:

erow
/

Medical-ROIs-K2.6

id stringlengths 15 19	source stringclasses 4 values	teacher_model stringclasses 1 value	teacher_provenance stringclasses 1 value	messages listlengths 2 2	messages_mm stringlengths 213 1.14k	problem stringlengths 12 270	solution stringlengths 9 650	prompt stringlengths 20 278	response stringlengths 64 963	seeing_regions listlengths 1 8	seeing_json stringlengths 60 833	reasoning stringlengths 1.8k 112k	n_images int64 1 1	index int64 0 13.6k
medix-rl-data_16535	RadGenome/PMC-VQA	moonshotai/Kimi-K2.6	distilled_from:moonshotai/Kimi-K2.6	[ { "role": "user", "content": "<image>\nWhat is the name of the surface shown in the micrographs?" }, { "role": "assistant", "content": "<region><title>Endothelial surface in panel A demonstrating confluent cellular pavement</title><bbox>[25,50,465,940]</bbox></region>\n<region><title>Endothelial...	[{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "\nWhat is the name of the surface shown in the micrographs?"}]}, {"role": "assistant", "content": "<region><title>Endothelial surface in panel A demonstrating confluent cellular pavement</title><bbox>[25,50,465,940]</bbox></region>\n<region><tit...	What is the name of the surface shown in the micrographs?	<MICROSCOPY> Endothelial surface	<image> What is the name of the surface shown in the micrographs?	<region><title>Endothelial surface in panel A demonstrating confluent cellular pavement</title><bbox>[25,50,465,940]</bbox></region> <region><title>Endothelial surface in panel B demonstrating aligned spindle-shaped cells</title><bbox>[535,50,965,940]</bbox></region>	[ { "title": "Endothelial surface in panel A demonstrating confluent cellular pavement", "bbox": [ 25, 50, 465, 940 ] }, { "title": "Endothelial surface in panel B demonstrating aligned spindle-shaped cells", "bbox": [ 535, 50, 965, 940 ] }...	{"regions": [{"title": "Endothelial surface in panel A demonstrating confluent cellular pavement", "bbox": [25, 50, 465, 940]}, {"title": "Endothelial surface in panel B demonstrating aligned spindle-shaped cells", "bbox": [535, 50, 965, 940]}]}	The user wants me to identify regions of interest (ROIs) in the provided micrograph image that evidence the answer "Endothelial surface" for the clinical question "What is the name of the surface shown in the micrographs?" Looking at the image: - It's a scanning electron micrograph (SEM) showing two panels, A and B. ...	1	2
medix-rl-data_48416	BoKelvin/SLAKE	moonshotai/Kimi-K2.6	distilled_from:moonshotai/Kimi-K2.6	[ { "role": "user", "content": "<image>\nWhat color does the right lung appear in the picture?" }, { "role": "assistant", "content": "<region><title>Right lung field appearing black due to air-filled parenchyma</title><bbox>[80,150,420,650]</bbox></region>" } ]	[{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "\nWhat color does the right lung appear in the picture?"}]}, {"role": "assistant", "content": "<region><title>Right lung field appearing black due to air-filled parenchyma</title><bbox>[80,150,420,650]</bbox></region>"}]	What color does the right lung appear in the picture?	<X_RAY> The right lung appears black in the picture.	<image> What color does the right lung appear in the picture?	<region><title>Right lung field appearing black due to air-filled parenchyma</title><bbox>[80,150,420,650]</bbox></region>	[ { "title": "Right lung field appearing black due to air-filled parenchyma", "bbox": [ 80, 150, 420, 650 ] } ]	{"regions": [{"title": "Right lung field appearing black due to air-filled parenchyma", "bbox": [80, 150, 420, 650]}]}	The user wants me to identify regions of interest (ROIs) in the chest X-ray image that support the answer to the question "What color does the right lung appear in the picture?" The gold answer is that the right lung appears black. In a chest X-ray, the lungs appear black (radiolucent) because they are filled with ai...	1	4
medix-rl-data_4270	flaviagiammarino/path-vqa	moonshotai/Kimi-K2.6	distilled_from:moonshotai/Kimi-K2.6	[ { "role": "user", "content": "<image>\nWhat medical condition is currently present?" }, { "role": "assistant", "content": "<region><title>Scattered petechiae upper right</title><bbox>[550,120,950,420]</bbox></region>\n<region><title>Dense petechiae mid-right</title><bbox>[600,400,980,700]</bbox>...	[{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "\nWhat medical condition is currently present?"}]}, {"role": "assistant", "content": "<region><title>Scattered petechiae upper right</title><bbox>[550,120,950,420]</bbox></region>\n<region><title>Dense petechiae mid-right</title><bbox>[600,400,9...	What medical condition is currently present?	<CLINICAL_PHOTOGRAPHY> The currently present medical condition is petechiae.	<image> What medical condition is currently present?	<region><title>Scattered petechiae upper right</title><bbox>[550,120,950,420]</bbox></region> <region><title>Dense petechiae mid-right</title><bbox>[600,400,980,700]</bbox></region> <region><title>Petechiae lower right</title><bbox>[450,700,850,960]</bbox></region> <region><title>Petechiae left-central</title><bbox>[20...	[ { "title": "Scattered petechiae upper right", "bbox": [ 550, 120, 950, 420 ] }, { "title": "Dense petechiae mid-right", "bbox": [ 600, 400, 980, 700 ] }, { "title": "Petechiae lower right", "bbox": [ 450, 700, ...	{"regions": [{"title": "Scattered petechiae upper right", "bbox": [550, 120, 950, 420]}, {"title": "Dense petechiae mid-right", "bbox": [600, 400, 980, 700]}, {"title": "Petechiae lower right", "bbox": [450, 700, 850, 960]}, {"title": "Petechiae left-central", "bbox": [200, 300, 500, 600]}]}	The user wants me to identify regions of interest (ROIs) in the image that evidence the diagnosis of petechiae. Looking at the image, it shows skin with numerous small red/purple pinpoint spots scattered across the surface. These are classic petechiae - small (1-2 mm), non-blanching, red or purple spots caused by mi...	1	5
medix-rl-data_42095	flaviagiammarino/path-vqa	moonshotai/Kimi-K2.6	distilled_from:moonshotai/Kimi-K2.6	[ { "role": "user", "content": "<image>\nWhat does this image show?" }, { "role": "assistant", "content": "<region><title>Large cavernous blood-filled vascular spaces in left dermis</title><bbox>[100,420,400,880]</bbox></region>\n<region><title>Dilated cavernous channels in mid-dermis</title><bbox...	[{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "\nWhat does this image show?"}]}, {"role": "assistant", "content": "<region><title>Large cavernous blood-filled vascular spaces in left dermis</title><bbox>[100,420,400,880]</bbox></region>\n<region><title>Dilated cavernous channels in mid-dermi...	What does this image show?	<MICROSCOPY> The image shows a typical cavernous lesion.	<image> What does this image show?	<region><title>Large cavernous blood-filled vascular spaces in left dermis</title><bbox>[100,420,400,880]</bbox></region> <region><title>Dilated cavernous channels in mid-dermis</title><bbox>[380,500,650,900]</bbox></region> <region><title>Cavernous vessels in right lower dermis</title><bbox>[620,750,880,960]</bbox></r...	[ { "title": "Large cavernous blood-filled vascular spaces in left dermis", "bbox": [ 100, 420, 400, 880 ] }, { "title": "Dilated cavernous channels in mid-dermis", "bbox": [ 380, 500, 650, 900 ] }, { "title": "Cavernous vessels in ri...	{"regions": [{"title": "Large cavernous blood-filled vascular spaces in left dermis", "bbox": [100, 420, 400, 880]}, {"title": "Dilated cavernous channels in mid-dermis", "bbox": [380, 500, 650, 900]}, {"title": "Cavernous vessels in right lower dermis", "bbox": [620, 750, 880, 960]}]}	The user wants me to identify regions of interest (ROIs) in the provided histopathology image that support the gold answer: "The image shows a typical cavernous lesion." Looking at the image, this appears to be a histopathology slide of skin showing a cavernous hemangioma (also known as a cavernous venous malformatio...	1	18
medix-rl-data_32573	RadGenome/PMC-VQA	moonshotai/Kimi-K2.6	distilled_from:moonshotai/Kimi-K2.6	[ { "role": "user", "content": "<image>\nWhat body part did the angiography examination examine?" }, { "role": "assistant", "content": "<region><title>Splenic artery and hilar branches in early arterial phase</title><bbox>[50,400,480,980]</bbox></region>\n<region><title>Intrasplenic vascular branc...	[{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "\nWhat body part did the angiography examination examine?"}]}, {"role": "assistant", "content": "<region><title>Splenic artery and hilar branches in early arterial phase</title><bbox>[50,400,480,980]</bbox></region>\n<region><title>Intrasplenic ...	What body part did the angiography examination examine?	<ANGIOGRAPHY> The spleen.	<image> What body part did the angiography examination examine?	<region><title>Splenic artery and hilar branches in early arterial phase</title><bbox>[50,400,480,980]</bbox></region> <region><title>Intrasplenic vascular branches and parenchymal blush in arterial phase</title><bbox>[180,100,500,450]</bbox></region> <region><title>Splenic parenchymal opacification in capillary-venous...	[ { "title": "Splenic artery and hilar branches in early arterial phase", "bbox": [ 50, 400, 480, 980 ] }, { "title": "Intrasplenic vascular branches and parenchymal blush in arterial phase", "bbox": [ 180, 100, 500, 450 ] }, { "title...	{"regions": [{"title": "Splenic artery and hilar branches in early arterial phase", "bbox": [50, 400, 480, 980]}, {"title": "Intrasplenic vascular branches and parenchymal blush in arterial phase", "bbox": [180, 100, 500, 450]}, {"title": "Splenic parenchymal opacification in capillary-venous phase", "bbox": [520, 50, ...	The user wants me to identify regions of interest in the provided medical image that support the answer "The spleen" to the question "What body part did the angiography examination examine?" Looking at the image: - It's a split image labeled "a" and "b" - Image (a) shows an angiogram with a catheter in what appears t...	1	9
medix-rl-data_9526	RadGenome/PMC-VQA	moonshotai/Kimi-K2.6	distilled_from:moonshotai/Kimi-K2.6	[ { "role": "user", "content": "<image>\nWhat does the green stain represent?" }, { "role": "assistant", "content": "<region><title>dMHC-positive regenerating fiber in panel (a)</title><bbox>[520,90,880,190]</bbox></region>\n<region><title>dMHC-positive regenerating fiber in panel (a)</title><bbox...	[{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "\nWhat does the green stain represent?"}]}, {"role": "assistant", "content": "<region><title>dMHC-positive regenerating fiber in panel (a)</title><bbox>[520,90,880,190]</bbox></region>\n<region><title>dMHC-positive regenerating fiber in panel (a...	What does the green stain represent?	<MICROSCOPY> Regeneration marker	<image> What does the green stain represent?	<region><title>dMHC-positive regenerating fiber in panel (a)</title><bbox>[520,90,880,190]</bbox></region> <region><title>dMHC-positive regenerating fiber in panel (a)</title><bbox>[280,260,480,330]</bbox></region> <region><title>Merged dMHC and desmin in regenerating fiber (panel c)</title><bbox>[560,760,840,830]</bbo...	[ { "title": "dMHC-positive regenerating fiber in panel (a)", "bbox": [ 520, 90, 880, 190 ] }, { "title": "dMHC-positive regenerating fiber in panel (a)", "bbox": [ 280, 260, 480, 330 ] }, { "title": "Merged dMHC and desmin in regener...	{"regions": [{"title": "dMHC-positive regenerating fiber in panel (a)", "bbox": [520, 90, 880, 190]}, {"title": "dMHC-positive regenerating fiber in panel (a)", "bbox": [280, 260, 480, 330]}, {"title": "Merged dMHC and desmin in regenerating fiber (panel c)", "bbox": [560, 760, 840, 830]}, {"title": "Merged dMHC and de...	The user wants me to identify regions of interest (ROIs) in the provided medical image that support the answer to the clinical question. Clinical question: "What does the green stain represent?" Gold answer: "Regeneration marker" Looking at the image: - It's a composite of three microscopy panels: (a) dMHC, (b) Desm...	1	17
medix-rl-data_27065	RadGenome/PMC-VQA	moonshotai/Kimi-K2.6	distilled_from:moonshotai/Kimi-K2.6	[ { "role": "user", "content": "<image>\nWhat is the name of the type of cancer shown in the image?" }, { "role": "assistant", "content": "<region><title>Infiltrating malignant epithelial nests</title><bbox>[30,40,360,380]</bbox></region>\n<region><title>Invasive tumor cords in desmoplastic stroma...	[{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "\nWhat is the name of the type of cancer shown in the image?"}]}, {"role": "assistant", "content": "<region><title>Infiltrating malignant epithelial nests</title><bbox>[30,40,360,380]</bbox></region>\n<region><title>Invasive tumor cords in desmo...	What is the name of the type of cancer shown in the image?	<MICROSCOPY> Breast Cancer	<image> What is the name of the type of cancer shown in the image?	<region><title>Infiltrating malignant epithelial nests</title><bbox>[30,40,360,380]</bbox></region> <region><title>Invasive tumor cords in desmoplastic stroma</title><bbox>[340,200,700,620]</bbox></region> <region><title>Malignant epithelial nests in fibrous stroma</title><bbox>[620,80,980,520]</bbox></region> <region>...	[ { "title": "Infiltrating malignant epithelial nests", "bbox": [ 30, 40, 360, 380 ] }, { "title": "Invasive tumor cords in desmoplastic stroma", "bbox": [ 340, 200, 700, 620 ] }, { "title": "Malignant epithelial nests in fibrous stro...	{"regions": [{"title": "Infiltrating malignant epithelial nests", "bbox": [30, 40, 360, 380]}, {"title": "Invasive tumor cords in desmoplastic stroma", "bbox": [340, 200, 700, 620]}, {"title": "Malignant epithelial nests in fibrous stroma", "bbox": [620, 80, 980, 520]}, {"title": "Infiltrative carcinoma nests", "bbox":...	The user wants me to identify regions of interest (ROIs) in a histopathology image that support the diagnosis of "Breast Cancer". Looking at the image: - It's an H&E stained histopathology slide. - The image shows infiltrating nests/clusters of malignant epithelial cells (dark purple basophilic areas) within a desmop...	1	15
medix-rl-data_27240	flaviagiammarino/path-vqa	moonshotai/Kimi-K2.6	distilled_from:moonshotai/Kimi-K2.6	[ { "role": "user", "content": "<image>\nIs there metastatic carcinoma present?" }, { "role": "assistant", "content": "<region><title>Metastatic carcinoma with central necrosis (upper)</title><bbox>[330,180,420,300]</bbox></region>\n<region><title>Pale infiltrative metastatic tumor (upper lobe)</t...	[{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "\nIs there metastatic carcinoma present?"}]}, {"role": "assistant", "content": "<region><title>Metastatic carcinoma with central necrosis (upper)</title><bbox>[330,180,420,300]</bbox></region>\n<region><title>Pale infiltrative metastatic tumor (...	Is there metastatic carcinoma present?	<CLINICAL_PHOTOGRAPHY> Yes, metastatic carcinoma is present.	<image> Is there metastatic carcinoma present?	<region><title>Metastatic carcinoma with central necrosis (upper)</title><bbox>[330,180,420,300]</bbox></region> <region><title>Pale infiltrative metastatic tumor (upper lobe)</title><bbox>[160,150,500,450]</bbox></region> <region><title>Metastatic carcinoma with central necrosis (lower)</title><bbox>[380,640,510,790]<...	[ { "title": "Metastatic carcinoma with central necrosis (upper)", "bbox": [ 330, 180, 420, 300 ] }, { "title": "Pale infiltrative metastatic tumor (upper lobe)", "bbox": [ 160, 150, 500, 450 ] }, { "title": "Metastatic carcinoma with...	{"regions": [{"title": "Metastatic carcinoma with central necrosis (upper)", "bbox": [330, 180, 420, 300]}, {"title": "Pale infiltrative metastatic tumor (upper lobe)", "bbox": [160, 150, 500, 450]}, {"title": "Metastatic carcinoma with central necrosis (lower)", "bbox": [380, 640, 510, 790]}, {"title": "Pale infiltrat...	We need to find ROIs that evidence metastatic carcinoma in this gross pathology image. The image shows a bisected organ (likely lymph node or ovary?) with multiple tan-white nodules and at least two distinct dark brown/black necrotic or hemorrhagic foci. The gold answer says "Yes, metastatic carcinoma is present." We ...	1	16
medix-rl-data_16911	BoKelvin/SLAKE	moonshotai/Kimi-K2.6	distilled_from:moonshotai/Kimi-K2.6	[ { "role": "user", "content": "<image>\nIs there a person named Brian in the picture?" }, { "role": "assistant", "content": "<region><title>Axial brain MRI scan showing the person who was imaged</title><bbox>[200,120,800,880]</bbox></region>" } ]	[{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "\nIs there a person named Brian in the picture?"}]}, {"role": "assistant", "content": "<region><title>Axial brain MRI scan showing the person who was imaged</title><bbox>[200,120,800,880]</bbox></region>"}]	Is there a person named Brian in the picture?	<MRI_SCAN> Yes, the picture contains a person named Brian.	<image> Is there a person named Brian in the picture?	<region><title>Axial brain MRI scan showing the person who was imaged</title><bbox>[200,120,800,880]</bbox></region>	[ { "title": "Axial brain MRI scan showing the person who was imaged", "bbox": [ 200, 120, 800, 880 ] } ]	{"regions": [{"title": "Axial brain MRI scan showing the person who was imaged", "bbox": [200, 120, 800, 880]}]}	The user wants me to identify ROIs in a medical image that support the answer to the clinical question: "Is there a person named Brian in the picture?" Wait, this is clearly a trick or test question. The image is an MRI scan of a brain (axial view, likely T2-weighted or FLAIR given the contrast). There is no text in ...	1	11
medix-rl-data_47271	RadGenome/PMC-VQA	moonshotai/Kimi-K2.6	distilled_from:moonshotai/Kimi-K2.6	[ { "role": "user", "content": "<image>\nWhat imaging modality is used in this diagnostic study?" }, { "role": "assistant", "content": "<region><title>L marker typical of plain film radiography</title><bbox>[820,40,870,90]</bbox></region>\n<region><title>Right clavicle and ribs showing radiopaque ...	[{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "\nWhat imaging modality is used in this diagnostic study?"}]}, {"role": "assistant", "content": "<region><title>L marker typical of plain film radiography</title><bbox>[820,40,870,90]</bbox></region>\n<region><title>Right clavicle and ribs showi...	What imaging modality is used in this diagnostic study?	<X_RAY> X-ray	<image> What imaging modality is used in this diagnostic study?	<region><title>L marker typical of plain film radiography</title><bbox>[820,40,870,90]</bbox></region> <region><title>Right clavicle and ribs showing radiopaque bony structures</title><bbox>[80,60,460,340]</bbox></region> <region><title>Left clavicle and ribs showing radiopaque bony structures</title><bbox>[540,60,920,...	[ { "title": "L marker typical of plain film radiography", "bbox": [ 820, 40, 870, 90 ] }, { "title": "Right clavicle and ribs showing radiopaque bony structures", "bbox": [ 80, 60, 460, 340 ] }, { "title": "Left clavicle and ribs sho...	{"regions": [{"title": "L marker typical of plain film radiography", "bbox": [820, 40, 870, 90]}, {"title": "Right clavicle and ribs showing radiopaque bony structures", "bbox": [80, 60, 460, 340]}, {"title": "Left clavicle and ribs showing radiopaque bony structures", "bbox": [540, 60, 920, 340]}, {"title": "Projected...	The user wants me to identify regions of interest (ROIs) in the provided medical image that evidence the answer to the clinical question. Clinical question: "What imaging modality is used in this diagnostic study?" Gold answer: "X-ray" I need to find visual features in the image that indicate this is an X-ray. Key f...	1	1
medix-rl-data_27367	flaviagiammarino/path-vqa	moonshotai/Kimi-K2.6	distilled_from:moonshotai/Kimi-K2.6	[ { "role": "user", "content": "<image>\nDoes the space between the layers of the pericardium contain numerous inflammatory cells, chiefly polymorphonuclear leukocytes (PMNs)?" }, { "role": "assistant", "content": "<region><title>Dense PMN infiltrate in pericardial space</title><bbox>[100,250,600,...	[{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "\nDoes the space between the layers of the pericardium contain numerous inflammatory cells, chiefly polymorphonuclear leukocytes (PMNs)?"}]}, {"role": "assistant", "content": "<region><title>Dense PMN infiltrate in pericardial space</title><bbox...	Does the space between the layers of the pericardium contain numerous inflammatory cells, chiefly polymorphonuclear leukocytes (PMNs)?	<MICROSCOPY> Yes, the space between the layers of the pericardium does contain numerous inflammatory cells, primarily polymorphonuclear leukocytes (PMNs).	<image> Does the space between the layers of the pericardium contain numerous inflammatory cells, chiefly polymorphonuclear leukocytes (PMNs)?	<region><title>Dense PMN infiltrate in pericardial space</title><bbox>[100,250,600,750]</bbox></region> <region><title>Pericardial layer with subjacent PMN-rich space</title><bbox>[0,700,400,1000]</bbox></region> <region><title>PMNs in space adjacent to fibrinous exudate</title><bbox>[500,200,950,700]</bbox></region>	[ { "title": "Dense PMN infiltrate in pericardial space", "bbox": [ 100, 250, 600, 750 ] }, { "title": "Pericardial layer with subjacent PMN-rich space", "bbox": [ 0, 700, 400, 1000 ] }, { "title": "PMNs in space adjacent to fibrinous...	{"regions": [{"title": "Dense PMN infiltrate in pericardial space", "bbox": [100, 250, 600, 750]}, {"title": "Pericardial layer with subjacent PMN-rich space", "bbox": [0, 700, 400, 1000]}, {"title": "PMNs in space adjacent to fibrinous exudate", "bbox": [500, 200, 950, 700]}]}	The user wants me to identify regions of interest (ROIs) in a histopathology image that support the answer to a clinical question. Clinical question: Does the space between the layers of the pericardium contain numerous inflammatory cells, chiefly polymorphonuclear leukocytes (PMNs)? Gold answer: Yes, the space betwe...	1	8
medix-rl-data_32959	flaviagiammarino/path-vqa	moonshotai/Kimi-K2.6	distilled_from:moonshotai/Kimi-K2.6	[ { "role": "user", "content": "<image>\nIs the image provided a good but not the best example of high new bone formation involving osteoblasts and osteoclasts?" }, { "role": "assistant", "content": "<region><title>Osteoclasts in left resorption pit</title><bbox>[130,140,400,580]</bbox></region>\n...	[{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "\nIs the image provided a good but not the best example of high new bone formation involving osteoblasts and osteoclasts?"}]}, {"role": "assistant", "content": "<region><title>Osteoclasts in left resorption pit</title><bbox>[130,140,400,580]</bb...	Is the image provided a good but not the best example of high new bone formation involving osteoblasts and osteoclasts?	<MICROSCOPY> Yes, the image is a good but not the best example of high new bone formation with osteoblasts and osteoclasts.	<image> Is the image provided a good but not the best example of high new bone formation involving osteoblasts and osteoclasts?	<region><title>Osteoclasts in left resorption pit</title><bbox>[130,140,400,580]</bbox></region> <region><title>Osteoclasts in right resorption pit</title><bbox>[490,290,870,720]</bbox></region> <region><title>Osteoblasts lining new bone surface</title><bbox>[180,650,550,970]</bbox></region> <region><title>Woven bone m...	[ { "title": "Osteoclasts in left resorption pit", "bbox": [ 130, 140, 400, 580 ] }, { "title": "Osteoclasts in right resorption pit", "bbox": [ 490, 290, 870, 720 ] }, { "title": "Osteoblasts lining new bone surface", "bbox": [ ...	{"regions": [{"title": "Osteoclasts in left resorption pit", "bbox": [130, 140, 400, 580]}, {"title": "Osteoclasts in right resorption pit", "bbox": [490, 290, 870, 720]}, {"title": "Osteoblasts lining new bone surface", "bbox": [180, 650, 550, 970]}, {"title": "Woven bone matrix with osteocytes", "bbox": [350, 50, 850...	The user wants me to identify regions of interest (ROIs) in a histopathology image that support the gold answer. The gold answer states that this is a good but not the best example of high new bone formation involving osteoblasts and osteoclasts. I need to look at the image and identify: 1. Areas of new bone formatio...	1	12
medix-rl-data_33297	flaviagiammarino/path-vqa	moonshotai/Kimi-K2.6	distilled_from:moonshotai/Kimi-K2.6	[ { "role": "user", "content": "<image>\nWhat is present in the human body?" }, { "role": "assistant", "content": "<region><title>Appendiceal mucosa with intestinal crypts (superior wall)</title><bbox>[0,0,1000,180]</bbox></region>\n<region><title>Appendiceal mucosa with intestinal crypts (left la...	[{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "\nWhat is present in the human body?"}]}, {"role": "assistant", "content": "<region><title>Appendiceal mucosa with intestinal crypts (superior wall)</title><bbox>[0,0,1000,180]</bbox></region>\n<region><title>Appendiceal mucosa with intestinal c...	What is present in the human body?	<MICROSCOPY> The appendix is present in the human body.	<image> What is present in the human body?	<region><title>Appendiceal mucosa with intestinal crypts (superior wall)</title><bbox>[0,0,1000,180]</bbox></region> <region><title>Appendiceal mucosa with intestinal crypts (left lateral wall)</title><bbox>[0,0,150,700]</bbox></region> <region><title>Appendiceal mucosa with intestinal crypts (right lateral wall)</titl...	[ { "title": "Appendiceal mucosa with intestinal crypts (superior wall)", "bbox": [ 0, 0, 1000, 180 ] }, { "title": "Appendiceal mucosa with intestinal crypts (left lateral wall)", "bbox": [ 0, 0, 150, 700 ] }, { "title": "Appendiceal...	{"regions": [{"title": "Appendiceal mucosa with intestinal crypts (superior wall)", "bbox": [0, 0, 1000, 180]}, {"title": "Appendiceal mucosa with intestinal crypts (left lateral wall)", "bbox": [0, 0, 150, 700]}, {"title": "Appendiceal mucosa with intestinal crypts (right lateral wall)", "bbox": [780, 0, 1000, 350]}]}	The user wants me to identify regions of interest (ROIs) in the provided medical image that support the gold answer. The image appears to be a histology slide showing a cross-section of the appendix. The gold answer states: "The appendix is present in the human body." Looking at the image, I can see: 1. The mucosal...	1	10

End of preview. Expand in Data Studio

Medical-ROIs-K2.6

Medical visual grounding SFT data: for each clinical VQA sample, a teacher model proposes answer-supporting ROIs (regions of interest) as 2D bounding boxes.

Teacher: moonshotai/Kimi-K2.6.
Upstream images & QA: MBZUAI/medix-rl-data.

How the data is generated

MBZUAI/medix-rl-data (train)
        │
        │  problem / solution / image / source / id
        ▼
  Teacher: moonshotai/Kimi-K2.6
  (vision + text; given question + gold answer)
        │
        │  ROI XML: <region><title>…</title><bbox>[x1,y1,x2,y2]</bbox></region>
        ▼
  JSONL (sft.jsonl)  →  stratified split
        │                 ├── sft_train.jsonl  (25,134)
        │                 └── sft_val.jsonl    (2,000)
        ▼
  prepare_hf: join JSONL text with medix-rl-data images
        │
        ▼
  Hugging Face DatasetDict  (this repo)

1. Source VQA pool

Rows come from MBZUAI/medix-rl-data (train, 51,335 examples) with four upstream sources:

source	role
`RadGenome/PMC-VQA`	PubMed Central figure VQA
`flaviagiammarino/path-vqa`	Pathology VQA
`BoKelvin/SLAKE`	Radiology VQA (SLAKE)
`flaviagiammarino/vqa-rad`	VQA-RAD

Each row provides id, image, problem (question), solution (gold answer), and source.

2. Teacher ROI annotation

A multimodal teacher (moonshotai/Kimi-K2.6) is prompted with the image, clinical question, and gold answer, and asked to box the visual evidence that supports the answer.

Output format (one or more blocks):

<region><title>caption</title><bbox>[x1,y1,x2,y2]</bbox></region>

BBox convention: integers in [0, 1000], top-left origin; require x1 < x2, y1 < y2.
Successful annotations are written to JSONL (sft.jsonl); failures are logged for repair/retry.
This release keeps the successfully annotated subset (27,134 rows, ≈53% of the upstream train pool).

3. Train / validation split

From sft.jsonl, a source-stratified random split (seed=42) produces:

split	file	count
train	`sft_train.jsonl`	25,134
validation	`sft_val.jsonl`	2,000

4. Hugging Face packaging

medix_seeing_sft.prepare_hf joins each JSONL row with the matching image(s) from MBZUAI/medix-rl-data by id, and materializes a DatasetDict with multimodal chat fields suitable for SFT loaders.

Splits in this repo

split	#examples
`train`	25,134
`validation`	2,000

Source mix (approx.)

source	train	validation
RadGenome/PMC-VQA	12,315	980
flaviagiammarino/path-vqa	9,535	759
BoKelvin/SLAKE	2,391	190
flaviagiammarino/vqa-rad	893	71

Schema

field	description
`id`	Sample id (matches `medix-rl-data`)
`source`	Upstream VQA dataset name
`images`	Image list (`datasets.Image`)
`messages`	Chat turns with string content (`<image>` placeholders)
`messages_mm`	JSON string of multimodal content-list messages
`problem` / `solution`	Clinical question and gold answer
`prompt` / `response`	User prompt and assistant ROI XML
`seeing_regions`	Structured `{title, bbox}` list
`seeing_json`	JSON serialization of regions
`reasoning`	Teacher reasoning text (when available)
`teacher_model`	`moonshotai/Kimi-K2.6`
`teacher_provenance`	`distilled_from:moonshotai/Kimi-K2.6`
`n_images`	Number of images
`index`	Annotation-run index

Example assistant target

<region><title>Right kidney</title><bbox>[130,380,470,720]</bbox></region>
<region><title>Left kidney</title><bbox>[570,420,840,750]</bbox></region>

Load

from datasets import load_dataset

ds = load_dataset("erow/Medical-ROIs-K2.6")
print(ds["train"][0]["problem"])
print(ds["train"][0]["response"])
ds["train"][0]["images"][0]  # PIL.Image

Or from a local save_to_disk folder:

from datasets import load_from_disk
ds = load_from_disk("/path/to/hf_seeing_sft")

Intended use

Supervised fine-tuning / distillation of medical VLMs for visual grounding: predicting ROIs that evidence the answer to a clinical question (complementary to answer-only VQA).

Limitations

ROIs are teacher-generated, not human-verified for every sample; some boxes may be loose, panel-misaligned, or weakly linked to the QA.
Coverage is a subset of medix-rl-data (annotation failures / rate limits excluded).
Not a clinical decision-support product; for research use only.

Citation / provenance

Upstream pool: MBZUAI/medix-rl-data and its constituent VQA datasets (PMC-VQA, PathVQA, SLAKE, VQA-RAD).
Teacher: Moonshot AI Kimi K2.6 (via OpenAI-compatible inference).
Pipeline: MediX seeing-SFT (medix_seeing_sft generate → split → prepare_hf).

Downloads last month: 69