Sign In

DeepTumorVQA: A Hierarchical 3D CT Benchmark for Stage-Wise Evaluation of Medical VLMs and Tool-Augmented Agents

์ž‘์„ฑ์ž
  • Haebom
์นดํ…Œ๊ณ ๋ฆฌ
Empty

์ €์ž

Yixiong Chen, Wenjie Xiao, Pedro R. A. S. Bassi, Boyan Wang, Liang He, Xinze Zhou, Sezgin Er, Ibrahim Ethem Hamamci, Zongwei Zhou, Alan Yuille

๐Ÿ’ก ๊ฐœ์š”

๋ณธ ์—ฐ๊ตฌ๋Š” 3D CT ์˜์ƒ ๊ธฐ๋ฐ˜ ์ข…์–‘ ์ง„๋‹จ ๊ณผ์ •์„ ๊ณ„์ธต์ ์œผ๋กœ ๋ถ„์„ํ•˜๊ณ  ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•œ ์ƒˆ๋กœ์šด ๋ฒค์น˜๋งˆํฌ์ธ DeepTumorVQA๋ฅผ ์ œ์•ˆํ•ฉ๋‹ˆ๋‹ค. ๊ธฐ์กด VQA ๋ฒค์น˜๋งˆํฌ์˜ ๋‹จ์ผ ์ •ํ™•๋„ ์ ์ˆ˜ ํ•œ๊ณ„๋ฅผ ๊ทน๋ณตํ•˜๊ณ , ์ธ์‹, ์ธก์ •, ์‹œ๊ฐ์  ์ถ”๋ก , ์˜ํ•™์  ์ถ”๋ก ์˜ ๋„ค ๋‹จ๊ณ„๋กœ ์ถ”๋ก  ๊ณผ์ •์„ ์„ธ๋ถ„ํ™”ํ•˜์—ฌ ๋ชจ๋ธ์˜ ์‹คํŒจ ์ง€์ ์„ ํŒŒ์•…ํ•  ์ˆ˜ ์žˆ๋„๋ก ํ•ฉ๋‹ˆ๋‹ค. 476,000๊ฐœ ์ด์ƒ์˜ ์งˆ๋ฌธ๊ณผ 9,262๊ฐœ์˜ 3D CT ๋ณผ๋ฅจ์„ ํฌํ•จํ•˜๋ฉฐ, ๋„๊ตฌ ์‚ฌ์šฉ ํ™˜๊ฒฝ๋„ ์ œ๊ณตํ•˜์—ฌ AI ์—์ด์ „ํŠธ ํ‰๊ฐ€๋ฅผ ์ง€์›ํ•ฉ๋‹ˆ๋‹ค.

๐Ÿ”‘ ์‹œ์‚ฌ์  ๋ฐ ํ•œ๊ณ„

โ€ข
์ •๋Ÿ‰์  ์ธก์ •์˜ ์ค‘์š”์„ฑ: ํ˜„์žฌ ์˜๋ฃŒ VLM์€ ์ •ํ™•ํ•œ ์ •๋Ÿ‰์  ์ธก์ •์ด ๊ฐ€์žฅ ํฐ ๋ณ‘๋ชฉ ํ˜„์ƒ์ด๋ฉฐ, ์ด๋Š” ํ›„๋ฐ˜ ๋‹จ๊ณ„์˜ ์ถ”๋ก  ๋Šฅ๋ ฅ์„ ์ €ํ•ดํ•˜๋Š” ์ฃผ์š” ์›์ธ์ž„์„ ํ™•์ธํ–ˆ์Šต๋‹ˆ๋‹ค.
โ€ข
๋„๊ตฌ ํ™œ์šฉ์˜ ํšจ๊ณผ: ์ด๋ฏธ์ง€ ๋ถ„ํ• , ์ธก์ •, ์˜๋ฃŒ ์ง€์‹ ๋ชจ๋“ˆ๊ณผ ๊ฐ™์€ ์™ธ๋ถ€ ๋„๊ตฌ ํ™œ์šฉ์ด VLM์˜ ์„ฑ๋Šฅ์„ ํฌ๊ฒŒ ํ–ฅ์ƒ์‹œํ‚ฌ ์ˆ˜ ์žˆ์Œ์„ ๋ณด์—ฌ์ฃผ์—ˆ์œผ๋ฉฐ, ์ด๋ฅผ ํ†ตํ•ด ์ƒˆ๋กœ์šด ๋„์ „ ๊ณผ์ œ๋กœ ์˜๋ฃŒ ์ง€์‹ ๋ฐ ๋„๊ตฌ๋ฅผ ํ™œ์šฉํ•œ ์ถ”๋ก ์ด ๋ถ€์ƒํ•จ์„ ์‹œ์‚ฌํ•ฉ๋‹ˆ๋‹ค.
โ€ข
๋ฐ์ดํ„ฐ ๋ฐ ์ฝ”๋“œ ๊ณต๊ฐœ: ๋ฒค์น˜๋งˆํฌ ๋ฐ์ดํ„ฐ์…‹๊ณผ ์ถ”๋ก ์„ ์œ„ํ•œ ๋„๊ตฌ ์‚ฌ์šฉ ํ”์ ์„ ๊ณต๊ฐœํ•˜์—ฌ ํ–ฅํ›„ ์˜๋ฃŒ VLM ๋ฐ AI ์—์ด์ „ํŠธ ์—ฐ๊ตฌ๋ฅผ ์œ„ํ•œ ๊ตฌ์ฒด์ ์ธ ๋กœ๋“œ๋งต์„ ์ œ๊ณตํ•˜๊ณ , ์ด๋ฅผ ํ†ตํ•ด ์—์ด์ „ํŠธ์˜ ๋„๊ตฌ ์‚ฌ์šฉ ๋ฐ ์ถ”๋ก  ์‹คํŒจ๋ฅผ ์ค„์ผ ์ˆ˜ ์žˆ๋Š” ์ง€๋„ ํ•™์Šต ํ™˜๊ฒฝ์„ ๊ตฌ์ถ•ํ•ฉ๋‹ˆ๋‹ค.
๐Ÿ‘
Made with Slashpage